2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_trans.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_alloc.h"
38 #include "xfs_bmap_util.h"
39 #include "xfs_bmap_btree.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_buf_item.h"
45 #include "xfs_trace.h"
46 #include "xfs_symlink.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_filestream.h"
50 #include "xfs_ag_resv.h"
51 #include "xfs_refcount.h"
52 #include "xfs_rmap_btree.h"
53 #include "xfs_icache.h"
56 kmem_zone_t
*xfs_bmap_free_item_zone
;
59 * Miscellaneous helper functions
63 * Compute and fill in the value of the maximum depth of a bmap btree
64 * in this filesystem. Done once, during mount.
67 xfs_bmap_compute_maxlevels(
68 xfs_mount_t
*mp
, /* file system mount structure */
69 int whichfork
) /* data or attr fork */
71 int level
; /* btree level */
72 uint maxblocks
; /* max blocks at this level */
73 uint maxleafents
; /* max leaf entries possible */
74 int maxrootrecs
; /* max records in root block */
75 int minleafrecs
; /* min records in leaf block */
76 int minnoderecs
; /* min records in node block */
77 int sz
; /* root block size */
80 * The maximum number of extents in a file, hence the maximum
81 * number of leaf entries, is controlled by the type of di_nextents
82 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
83 * (a signed 16-bit number, xfs_aextnum_t).
85 * Note that we can no longer assume that if we are in ATTR1 that
86 * the fork offset of all the inodes will be
87 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
88 * with ATTR2 and then mounted back with ATTR1, keeping the
89 * di_forkoff's fixed but probably at various positions. Therefore,
90 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
91 * of a minimum size available.
93 if (whichfork
== XFS_DATA_FORK
) {
94 maxleafents
= MAXEXTNUM
;
95 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
97 maxleafents
= MAXAEXTNUM
;
98 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
100 maxrootrecs
= xfs_bmdr_maxrecs(sz
, 0);
101 minleafrecs
= mp
->m_bmap_dmnr
[0];
102 minnoderecs
= mp
->m_bmap_dmnr
[1];
103 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
104 for (level
= 1; maxblocks
> 1; level
++) {
105 if (maxblocks
<= maxrootrecs
)
108 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
110 mp
->m_bm_maxlevels
[whichfork
] = level
;
113 STATIC
int /* error */
115 struct xfs_btree_cur
*cur
,
119 int *stat
) /* success/failure */
121 cur
->bc_rec
.b
.br_startoff
= off
;
122 cur
->bc_rec
.b
.br_startblock
= bno
;
123 cur
->bc_rec
.b
.br_blockcount
= len
;
124 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
127 STATIC
int /* error */
129 struct xfs_btree_cur
*cur
,
133 int *stat
) /* success/failure */
135 cur
->bc_rec
.b
.br_startoff
= off
;
136 cur
->bc_rec
.b
.br_startblock
= bno
;
137 cur
->bc_rec
.b
.br_blockcount
= len
;
138 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
142 * Check if the inode needs to be converted to btree format.
144 static inline bool xfs_bmap_needs_btree(struct xfs_inode
*ip
, int whichfork
)
146 return whichfork
!= XFS_COW_FORK
&&
147 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
148 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
149 XFS_IFORK_MAXEXT(ip
, whichfork
);
153 * Check if the inode should be converted to extent format.
155 static inline bool xfs_bmap_wants_extents(struct xfs_inode
*ip
, int whichfork
)
157 return whichfork
!= XFS_COW_FORK
&&
158 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
&&
159 XFS_IFORK_NEXTENTS(ip
, whichfork
) <=
160 XFS_IFORK_MAXEXT(ip
, whichfork
);
164 * Update the record referred to by cur to the value given
165 * by [off, bno, len, state].
166 * This either works (return 0) or gets an EFSCORRUPTED error.
170 struct xfs_btree_cur
*cur
,
176 union xfs_btree_rec rec
;
178 xfs_bmbt_disk_set_allf(&rec
.bmbt
, off
, bno
, len
, state
);
179 return xfs_btree_update(cur
, &rec
);
183 * Compute the worst-case number of indirect blocks that will be used
184 * for ip's delayed extent of length "len".
187 xfs_bmap_worst_indlen(
188 xfs_inode_t
*ip
, /* incore inode pointer */
189 xfs_filblks_t len
) /* delayed extent length */
191 int level
; /* btree level number */
192 int maxrecs
; /* maximum record count at this level */
193 xfs_mount_t
*mp
; /* mount structure */
194 xfs_filblks_t rval
; /* return value */
195 xfs_filblks_t orig_len
;
199 /* Calculate the worst-case size of the bmbt. */
201 maxrecs
= mp
->m_bmap_dmxr
[0];
202 for (level
= 0, rval
= 0;
203 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
206 do_div(len
, maxrecs
);
209 rval
+= XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
214 maxrecs
= mp
->m_bmap_dmxr
[1];
217 /* Calculate the worst-case size of the rmapbt. */
218 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
))
219 rval
+= 1 + xfs_rmapbt_calc_size(mp
, orig_len
) +
220 mp
->m_rmap_maxlevels
;
226 * Calculate the default attribute fork offset for newly created inodes.
229 xfs_default_attroffset(
230 struct xfs_inode
*ip
)
232 struct xfs_mount
*mp
= ip
->i_mount
;
235 if (mp
->m_sb
.sb_inodesize
== 256) {
236 offset
= XFS_LITINO(mp
, ip
->i_d
.di_version
) -
237 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
239 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
242 ASSERT(offset
< XFS_LITINO(mp
, ip
->i_d
.di_version
));
247 * Helper routine to reset inode di_forkoff field when switching
248 * attribute fork from local to extent format - we reset it where
249 * possible to make space available for inline data fork extents.
252 xfs_bmap_forkoff_reset(
256 if (whichfork
== XFS_ATTR_FORK
&&
257 ip
->i_d
.di_format
!= XFS_DINODE_FMT_DEV
&&
258 ip
->i_d
.di_format
!= XFS_DINODE_FMT_UUID
&&
259 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) {
260 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
262 if (dfl_forkoff
> ip
->i_d
.di_forkoff
)
263 ip
->i_d
.di_forkoff
= dfl_forkoff
;
268 STATIC
struct xfs_buf
*
270 struct xfs_btree_cur
*cur
,
273 struct xfs_log_item_desc
*lidp
;
279 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
280 if (!cur
->bc_bufs
[i
])
282 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
283 return cur
->bc_bufs
[i
];
286 /* Chase down all the log items to see if the bp is there */
287 list_for_each_entry(lidp
, &cur
->bc_tp
->t_items
, lid_trans
) {
288 struct xfs_buf_log_item
*bip
;
289 bip
= (struct xfs_buf_log_item
*)lidp
->lid_item
;
290 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
291 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
300 struct xfs_btree_block
*block
,
306 __be64
*pp
, *thispa
; /* pointer to block address */
307 xfs_bmbt_key_t
*prevp
, *keyp
;
309 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
312 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
313 dmxr
= mp
->m_bmap_dmxr
[0];
314 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
317 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
318 be64_to_cpu(keyp
->br_startoff
));
323 * Compare the block numbers to see if there are dups.
326 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
328 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
330 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
332 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
334 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
335 if (*thispa
== *pp
) {
336 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
338 (unsigned long long)be64_to_cpu(*thispa
));
339 panic("%s: ptrs are equal in node\n",
347 * Check that the extents for the inode ip are in the right order in all
348 * btree leaves. THis becomes prohibitively expensive for large extent count
349 * files, so don't bother with inodes that have more than 10,000 extents in
350 * them. The btree record ordering checks will still be done, so for such large
351 * bmapbt constructs that is going to catch most corruptions.
354 xfs_bmap_check_leaf_extents(
355 xfs_btree_cur_t
*cur
, /* btree cursor or null */
356 xfs_inode_t
*ip
, /* incore inode pointer */
357 int whichfork
) /* data or attr fork */
359 struct xfs_btree_block
*block
; /* current btree block */
360 xfs_fsblock_t bno
; /* block # of "block" */
361 xfs_buf_t
*bp
; /* buffer for "block" */
362 int error
; /* error return value */
363 xfs_extnum_t i
=0, j
; /* index into the extents list */
364 xfs_ifork_t
*ifp
; /* fork structure */
365 int level
; /* btree level, for checking */
366 xfs_mount_t
*mp
; /* file system mount structure */
367 __be64
*pp
; /* pointer to block address */
368 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
369 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
370 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
373 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
) {
377 /* skip large extent count inodes */
378 if (ip
->i_d
.di_nextents
> 10000)
383 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
384 block
= ifp
->if_broot
;
386 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
388 level
= be16_to_cpu(block
->bb_level
);
390 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
391 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
392 bno
= be64_to_cpu(*pp
);
394 ASSERT(bno
!= NULLFSBLOCK
);
395 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
396 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
399 * Go down the tree until leaf level is reached, following the first
400 * pointer (leftmost) at each level.
402 while (level
-- > 0) {
403 /* See if buf is in cur first */
405 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
408 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
414 block
= XFS_BUF_TO_BLOCK(bp
);
419 * Check this block for basic sanity (increasing keys and
420 * no duplicate blocks).
423 xfs_check_block(block
, mp
, 0, 0);
424 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
425 bno
= be64_to_cpu(*pp
);
426 XFS_WANT_CORRUPTED_GOTO(mp
,
427 XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
430 xfs_trans_brelse(NULL
, bp
);
435 * Here with bp and block set to the leftmost leaf node in the tree.
440 * Loop over all leaf nodes checking that all extents are in the right order.
443 xfs_fsblock_t nextbno
;
444 xfs_extnum_t num_recs
;
447 num_recs
= xfs_btree_get_numrecs(block
);
450 * Read-ahead the next leaf block, if any.
453 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
456 * Check all the extents to make sure they are OK.
457 * If we had a previous block, the last entry should
458 * conform with the first entry in this one.
461 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
463 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
464 xfs_bmbt_disk_get_blockcount(&last
) <=
465 xfs_bmbt_disk_get_startoff(ep
));
467 for (j
= 1; j
< num_recs
; j
++) {
468 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
469 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
470 xfs_bmbt_disk_get_blockcount(ep
) <=
471 xfs_bmbt_disk_get_startoff(nextp
));
479 xfs_trans_brelse(NULL
, bp
);
483 * If we've reached the end, stop.
485 if (bno
== NULLFSBLOCK
)
489 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
492 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
498 block
= XFS_BUF_TO_BLOCK(bp
);
504 xfs_warn(mp
, "%s: at error0", __func__
);
506 xfs_trans_brelse(NULL
, bp
);
508 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
510 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__
);
515 * Add bmap trace insert entries for all the contents of the extent records.
518 xfs_bmap_trace_exlist(
519 xfs_inode_t
*ip
, /* incore inode pointer */
520 xfs_extnum_t cnt
, /* count of entries in the list */
521 int whichfork
, /* data or attr or cow fork */
522 unsigned long caller_ip
)
524 xfs_extnum_t idx
; /* extent record index */
525 xfs_ifork_t
*ifp
; /* inode fork pointer */
528 if (whichfork
== XFS_ATTR_FORK
)
529 state
|= BMAP_ATTRFORK
;
530 else if (whichfork
== XFS_COW_FORK
)
531 state
|= BMAP_COWFORK
;
533 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
534 ASSERT(cnt
== xfs_iext_count(ifp
));
535 for (idx
= 0; idx
< cnt
; idx
++)
536 trace_xfs_extlist(ip
, idx
, state
, caller_ip
);
540 * Validate that the bmbt_irecs being returned from bmapi are valid
541 * given the caller's original parameters. Specifically check the
542 * ranges of the returned irecs to ensure that they only extend beyond
543 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
546 xfs_bmap_validate_ret(
550 xfs_bmbt_irec_t
*mval
,
554 int i
; /* index to map values */
556 ASSERT(ret_nmap
<= nmap
);
558 for (i
= 0; i
< ret_nmap
; i
++) {
559 ASSERT(mval
[i
].br_blockcount
> 0);
560 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
561 ASSERT(mval
[i
].br_startoff
>= bno
);
562 ASSERT(mval
[i
].br_blockcount
<= len
);
563 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
566 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
567 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
571 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
572 mval
[i
].br_startoff
);
573 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
574 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
575 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
576 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
581 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
582 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
586 * bmap free list manipulation functions
590 * Add the extent to the list of extents to be free at transaction end.
591 * The list is maintained sorted (by block number).
595 struct xfs_mount
*mp
,
596 struct xfs_defer_ops
*dfops
,
599 struct xfs_owner_info
*oinfo
)
601 struct xfs_extent_free_item
*new; /* new element */
606 ASSERT(bno
!= NULLFSBLOCK
);
608 ASSERT(len
<= MAXEXTLEN
);
609 ASSERT(!isnullstartblock(bno
));
610 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
611 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
612 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
613 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
614 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
615 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
617 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
619 new = kmem_zone_alloc(xfs_bmap_free_item_zone
, KM_SLEEP
);
620 new->xefi_startblock
= bno
;
621 new->xefi_blockcount
= (xfs_extlen_t
)len
;
623 new->xefi_oinfo
= *oinfo
;
625 xfs_rmap_skip_owner_update(&new->xefi_oinfo
);
626 trace_xfs_bmap_free_defer(mp
, XFS_FSB_TO_AGNO(mp
, bno
), 0,
627 XFS_FSB_TO_AGBNO(mp
, bno
), len
);
628 xfs_defer_add(dfops
, XFS_DEFER_OPS_TYPE_FREE
, &new->xefi_list
);
632 * Inode fork format manipulation functions
636 * Transform a btree format file with only one leaf node, where the
637 * extents list will fit in the inode, into an extents format file.
638 * Since the file extents are already in-core, all we have to do is
639 * give up the space for the btree root and pitch the leaf block.
641 STATIC
int /* error */
642 xfs_bmap_btree_to_extents(
643 xfs_trans_t
*tp
, /* transaction pointer */
644 xfs_inode_t
*ip
, /* incore inode pointer */
645 xfs_btree_cur_t
*cur
, /* btree cursor */
646 int *logflagsp
, /* inode logging flags */
647 int whichfork
) /* data or attr fork */
650 struct xfs_btree_block
*cblock
;/* child btree block */
651 xfs_fsblock_t cbno
; /* child block number */
652 xfs_buf_t
*cbp
; /* child block's buffer */
653 int error
; /* error return value */
654 xfs_ifork_t
*ifp
; /* inode fork data */
655 xfs_mount_t
*mp
; /* mount point structure */
656 __be64
*pp
; /* ptr to block address */
657 struct xfs_btree_block
*rblock
;/* root btree block */
658 struct xfs_owner_info oinfo
;
661 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
662 ASSERT(whichfork
!= XFS_COW_FORK
);
663 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
664 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
665 rblock
= ifp
->if_broot
;
666 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
667 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
668 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
669 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
670 cbno
= be64_to_cpu(*pp
);
673 if ((error
= xfs_btree_check_lptr(cur
, cbno
, 1)))
676 error
= xfs_btree_read_bufl(mp
, tp
, cbno
, 0, &cbp
, XFS_BMAP_BTREE_REF
,
680 cblock
= XFS_BUF_TO_BLOCK(cbp
);
681 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
683 xfs_rmap_ino_bmbt_owner(&oinfo
, ip
->i_ino
, whichfork
);
684 xfs_bmap_add_free(mp
, cur
->bc_private
.b
.dfops
, cbno
, 1, &oinfo
);
685 ip
->i_d
.di_nblocks
--;
686 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
687 xfs_trans_binval(tp
, cbp
);
688 if (cur
->bc_bufs
[0] == cbp
)
689 cur
->bc_bufs
[0] = NULL
;
690 xfs_iroot_realloc(ip
, -1, whichfork
);
691 ASSERT(ifp
->if_broot
== NULL
);
692 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
693 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
694 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
699 * Convert an extents-format file into a btree-format file.
700 * The new file will have a root block (in the inode) and a single child block.
702 STATIC
int /* error */
703 xfs_bmap_extents_to_btree(
704 xfs_trans_t
*tp
, /* transaction pointer */
705 xfs_inode_t
*ip
, /* incore inode pointer */
706 xfs_fsblock_t
*firstblock
, /* first-block-allocated */
707 struct xfs_defer_ops
*dfops
, /* blocks freed in xaction */
708 xfs_btree_cur_t
**curp
, /* cursor returned to caller */
709 int wasdel
, /* converting a delayed alloc */
710 int *logflagsp
, /* inode logging flags */
711 int whichfork
) /* data or attr fork */
713 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
714 xfs_buf_t
*abp
; /* buffer for ablock */
715 xfs_alloc_arg_t args
; /* allocation arguments */
716 xfs_bmbt_rec_t
*arp
; /* child record pointer */
717 struct xfs_btree_block
*block
; /* btree root block */
718 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
719 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
720 int error
; /* error return value */
721 xfs_extnum_t i
, cnt
; /* extent record index */
722 xfs_ifork_t
*ifp
; /* inode fork pointer */
723 xfs_bmbt_key_t
*kp
; /* root block key pointer */
724 xfs_mount_t
*mp
; /* mount structure */
725 xfs_extnum_t nextents
; /* number of file extents */
726 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
729 ASSERT(whichfork
!= XFS_COW_FORK
);
730 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
731 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
);
734 * Make space in the inode incore.
736 xfs_iroot_realloc(ip
, 1, whichfork
);
737 ifp
->if_flags
|= XFS_IFBROOT
;
742 block
= ifp
->if_broot
;
743 if (xfs_sb_version_hascrc(&mp
->m_sb
))
744 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
745 XFS_BMAP_CRC_MAGIC
, 1, 1, ip
->i_ino
,
746 XFS_BTREE_LONG_PTRS
| XFS_BTREE_CRC_BLOCKS
);
748 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
749 XFS_BMAP_MAGIC
, 1, 1, ip
->i_ino
,
750 XFS_BTREE_LONG_PTRS
);
753 * Need a cursor. Can't allocate until bb_level is filled in.
755 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
756 cur
->bc_private
.b
.firstblock
= *firstblock
;
757 cur
->bc_private
.b
.dfops
= dfops
;
758 cur
->bc_private
.b
.flags
= wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
760 * Convert to a btree with two levels, one record in root.
762 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_BTREE
);
763 memset(&args
, 0, sizeof(args
));
766 xfs_rmap_ino_bmbt_owner(&args
.oinfo
, ip
->i_ino
, whichfork
);
767 args
.firstblock
= *firstblock
;
768 if (*firstblock
== NULLFSBLOCK
) {
769 args
.type
= XFS_ALLOCTYPE_START_BNO
;
770 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
771 } else if (dfops
->dop_low
) {
773 args
.type
= XFS_ALLOCTYPE_START_BNO
;
774 args
.fsbno
= *firstblock
;
776 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
777 args
.fsbno
= *firstblock
;
779 args
.minlen
= args
.maxlen
= args
.prod
= 1;
780 args
.wasdel
= wasdel
;
782 if ((error
= xfs_alloc_vextent(&args
))) {
783 xfs_iroot_realloc(ip
, -1, whichfork
);
784 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
789 * During a CoW operation, the allocation and bmbt updates occur in
790 * different transactions. The mapping code tries to put new bmbt
791 * blocks near extents being mapped, but the only way to guarantee this
792 * is if the alloc and the mapping happen in a single transaction that
793 * has a block reservation. That isn't the case here, so if we run out
794 * of space we'll try again with another AG.
796 if (xfs_sb_version_hasreflink(&cur
->bc_mp
->m_sb
) &&
797 args
.fsbno
== NULLFSBLOCK
&&
798 args
.type
== XFS_ALLOCTYPE_NEAR_BNO
) {
799 dfops
->dop_low
= true;
803 * Allocation can't fail, the space was reserved.
805 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
806 ASSERT(*firstblock
== NULLFSBLOCK
||
807 args
.agno
== XFS_FSB_TO_AGNO(mp
, *firstblock
) ||
809 args
.agno
> XFS_FSB_TO_AGNO(mp
, *firstblock
)));
810 *firstblock
= cur
->bc_private
.b
.firstblock
= args
.fsbno
;
811 cur
->bc_private
.b
.allocated
++;
812 ip
->i_d
.di_nblocks
++;
813 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
814 abp
= xfs_btree_get_bufl(mp
, tp
, args
.fsbno
, 0);
816 * Fill in the child block.
818 abp
->b_ops
= &xfs_bmbt_buf_ops
;
819 ablock
= XFS_BUF_TO_BLOCK(abp
);
820 if (xfs_sb_version_hascrc(&mp
->m_sb
))
821 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
822 XFS_BMAP_CRC_MAGIC
, 0, 0, ip
->i_ino
,
823 XFS_BTREE_LONG_PTRS
| XFS_BTREE_CRC_BLOCKS
);
825 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
826 XFS_BMAP_MAGIC
, 0, 0, ip
->i_ino
,
827 XFS_BTREE_LONG_PTRS
);
829 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
830 nextents
= xfs_iext_count(ifp
);
831 for (cnt
= i
= 0; i
< nextents
; i
++) {
832 ep
= xfs_iext_get_ext(ifp
, i
);
833 if (!isnullstartblock(xfs_bmbt_get_startblock(ep
))) {
834 arp
->l0
= cpu_to_be64(ep
->l0
);
835 arp
->l1
= cpu_to_be64(ep
->l1
);
839 ASSERT(cnt
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
840 xfs_btree_set_numrecs(ablock
, cnt
);
843 * Fill in the root key and pointer.
845 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
846 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
847 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
848 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
849 be16_to_cpu(block
->bb_level
)));
850 *pp
= cpu_to_be64(args
.fsbno
);
853 * Do all this logging at the end so that
854 * the root is at the right level.
856 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
857 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
858 ASSERT(*curp
== NULL
);
860 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
865 * Convert a local file to an extents file.
866 * This code is out of bounds for data forks of regular files,
867 * since the file data needs to get logged so things will stay consistent.
868 * (The bmap-level manipulations are ok, though).
871 xfs_bmap_local_to_extents_empty(
872 struct xfs_inode
*ip
,
875 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
877 ASSERT(whichfork
!= XFS_COW_FORK
);
878 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
879 ASSERT(ifp
->if_bytes
== 0);
880 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0);
882 xfs_bmap_forkoff_reset(ip
, whichfork
);
883 ifp
->if_flags
&= ~XFS_IFINLINE
;
884 ifp
->if_flags
|= XFS_IFEXTENTS
;
885 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
889 STATIC
int /* error */
890 xfs_bmap_local_to_extents(
891 xfs_trans_t
*tp
, /* transaction pointer */
892 xfs_inode_t
*ip
, /* incore inode pointer */
893 xfs_fsblock_t
*firstblock
, /* first block allocated in xaction */
894 xfs_extlen_t total
, /* total blocks needed by transaction */
895 int *logflagsp
, /* inode logging flags */
897 void (*init_fn
)(struct xfs_trans
*tp
,
899 struct xfs_inode
*ip
,
900 struct xfs_ifork
*ifp
))
903 int flags
; /* logging flags returned */
904 xfs_ifork_t
*ifp
; /* inode fork pointer */
905 xfs_alloc_arg_t args
; /* allocation arguments */
906 xfs_buf_t
*bp
; /* buffer for extent block */
907 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
910 * We don't want to deal with the case of keeping inode data inline yet.
911 * So sending the data fork of a regular inode is invalid.
913 ASSERT(!(S_ISREG(VFS_I(ip
)->i_mode
) && whichfork
== XFS_DATA_FORK
));
914 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
915 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
917 if (!ifp
->if_bytes
) {
918 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
919 flags
= XFS_ILOG_CORE
;
925 ASSERT((ifp
->if_flags
& (XFS_IFINLINE
|XFS_IFEXTENTS
|XFS_IFEXTIREC
)) ==
927 memset(&args
, 0, sizeof(args
));
929 args
.mp
= ip
->i_mount
;
930 xfs_rmap_ino_owner(&args
.oinfo
, ip
->i_ino
, whichfork
, 0);
931 args
.firstblock
= *firstblock
;
933 * Allocate a block. We know we need only one, since the
934 * file currently fits in an inode.
936 if (*firstblock
== NULLFSBLOCK
) {
938 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
939 args
.type
= XFS_ALLOCTYPE_START_BNO
;
941 args
.fsbno
= *firstblock
;
942 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
945 args
.minlen
= args
.maxlen
= args
.prod
= 1;
946 error
= xfs_alloc_vextent(&args
);
951 * During a CoW operation, the allocation and bmbt updates occur in
952 * different transactions. The mapping code tries to put new bmbt
953 * blocks near extents being mapped, but the only way to guarantee this
954 * is if the alloc and the mapping happen in a single transaction that
955 * has a block reservation. That isn't the case here, so if we run out
956 * of space we'll try again with another AG.
958 if (xfs_sb_version_hasreflink(&ip
->i_mount
->m_sb
) &&
959 args
.fsbno
== NULLFSBLOCK
&&
960 args
.type
== XFS_ALLOCTYPE_NEAR_BNO
) {
963 /* Can't fail, the space was reserved. */
964 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
965 ASSERT(args
.len
== 1);
966 *firstblock
= args
.fsbno
;
967 bp
= xfs_btree_get_bufl(args
.mp
, tp
, args
.fsbno
, 0);
970 * Initialize the block, copy the data and log the remote buffer.
972 * The callout is responsible for logging because the remote format
973 * might differ from the local format and thus we don't know how much to
974 * log here. Note that init_fn must also set the buffer log item type
977 init_fn(tp
, bp
, ip
, ifp
);
979 /* account for the change in fork size */
980 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
981 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
982 flags
|= XFS_ILOG_CORE
;
984 xfs_iext_add(ifp
, 0, 1);
985 ep
= xfs_iext_get_ext(ifp
, 0);
986 xfs_bmbt_set_allf(ep
, 0, args
.fsbno
, 1, XFS_EXT_NORM
);
987 trace_xfs_bmap_post_update(ip
, 0,
988 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0,
990 XFS_IFORK_NEXT_SET(ip
, whichfork
, 1);
991 ip
->i_d
.di_nblocks
= 1;
992 xfs_trans_mod_dquot_byino(tp
, ip
,
993 XFS_TRANS_DQ_BCOUNT
, 1L);
994 flags
|= xfs_ilog_fext(whichfork
);
1002 * Called from xfs_bmap_add_attrfork to handle btree format files.
1004 STATIC
int /* error */
1005 xfs_bmap_add_attrfork_btree(
1006 xfs_trans_t
*tp
, /* transaction pointer */
1007 xfs_inode_t
*ip
, /* incore inode pointer */
1008 xfs_fsblock_t
*firstblock
, /* first block allocated */
1009 struct xfs_defer_ops
*dfops
, /* blocks to free at commit */
1010 int *flags
) /* inode logging flags */
1012 xfs_btree_cur_t
*cur
; /* btree cursor */
1013 int error
; /* error return value */
1014 xfs_mount_t
*mp
; /* file system mount struct */
1015 int stat
; /* newroot status */
1018 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
1019 *flags
|= XFS_ILOG_DBROOT
;
1021 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
1022 cur
->bc_private
.b
.dfops
= dfops
;
1023 cur
->bc_private
.b
.firstblock
= *firstblock
;
1024 if ((error
= xfs_bmbt_lookup_ge(cur
, 0, 0, 0, &stat
)))
1026 /* must be at least one entry */
1027 XFS_WANT_CORRUPTED_GOTO(mp
, stat
== 1, error0
);
1028 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
1031 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1034 *firstblock
= cur
->bc_private
.b
.firstblock
;
1035 cur
->bc_private
.b
.allocated
= 0;
1036 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1040 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1045 * Called from xfs_bmap_add_attrfork to handle extents format files.
1047 STATIC
int /* error */
1048 xfs_bmap_add_attrfork_extents(
1049 xfs_trans_t
*tp
, /* transaction pointer */
1050 xfs_inode_t
*ip
, /* incore inode pointer */
1051 xfs_fsblock_t
*firstblock
, /* first block allocated */
1052 struct xfs_defer_ops
*dfops
, /* blocks to free at commit */
1053 int *flags
) /* inode logging flags */
1055 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
1056 int error
; /* error return value */
1058 if (ip
->i_d
.di_nextents
* sizeof(xfs_bmbt_rec_t
) <= XFS_IFORK_DSIZE(ip
))
1061 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, dfops
, &cur
, 0,
1062 flags
, XFS_DATA_FORK
);
1064 cur
->bc_private
.b
.allocated
= 0;
1065 xfs_btree_del_cursor(cur
,
1066 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
1072 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1073 * different data fork content type needs a different callout to do the
1074 * conversion. Some are basic and only require special block initialisation
1075 * callouts for the data formating, others (directories) are so specialised they
1076 * handle everything themselves.
1078 * XXX (dgc): investigate whether directory conversion can use the generic
1079 * formatting callout. It should be possible - it's just a very complex
1082 STATIC
int /* error */
1083 xfs_bmap_add_attrfork_local(
1084 xfs_trans_t
*tp
, /* transaction pointer */
1085 xfs_inode_t
*ip
, /* incore inode pointer */
1086 xfs_fsblock_t
*firstblock
, /* first block allocated */
1087 struct xfs_defer_ops
*dfops
, /* blocks to free at commit */
1088 int *flags
) /* inode logging flags */
1090 xfs_da_args_t dargs
; /* args for dir/attr code */
1092 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
1095 if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
1096 memset(&dargs
, 0, sizeof(dargs
));
1097 dargs
.geo
= ip
->i_mount
->m_dir_geo
;
1099 dargs
.firstblock
= firstblock
;
1100 dargs
.dfops
= dfops
;
1101 dargs
.total
= dargs
.geo
->fsbcount
;
1102 dargs
.whichfork
= XFS_DATA_FORK
;
1104 return xfs_dir2_sf_to_block(&dargs
);
1107 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1108 return xfs_bmap_local_to_extents(tp
, ip
, firstblock
, 1,
1109 flags
, XFS_DATA_FORK
,
1110 xfs_symlink_local_to_remote
);
1112 /* should only be called for types that support local format data */
1114 return -EFSCORRUPTED
;
1118 * Convert inode from non-attributed to attributed.
1119 * Must not be in a transaction, ip must not be locked.
1121 int /* error code */
1122 xfs_bmap_add_attrfork(
1123 xfs_inode_t
*ip
, /* incore inode pointer */
1124 int size
, /* space new attribute needs */
1125 int rsvd
) /* xact may use reserved blks */
1127 xfs_fsblock_t firstblock
; /* 1st block/ag allocated */
1128 struct xfs_defer_ops dfops
; /* freed extent records */
1129 xfs_mount_t
*mp
; /* mount structure */
1130 xfs_trans_t
*tp
; /* transaction pointer */
1131 int blks
; /* space reservation */
1132 int version
= 1; /* superblock attr version */
1133 int logflags
; /* logging flags */
1134 int error
; /* error return value */
1136 ASSERT(XFS_IFORK_Q(ip
) == 0);
1139 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1141 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
1143 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_addafork
, blks
, 0,
1144 rsvd
? XFS_TRANS_RESERVE
: 0, &tp
);
1148 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1149 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
1150 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
1151 XFS_QMOPT_RES_REGBLKS
);
1154 if (XFS_IFORK_Q(ip
))
1156 if (ip
->i_d
.di_anextents
!= 0) {
1157 error
= -EFSCORRUPTED
;
1160 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
) {
1162 * For inodes coming from pre-6.2 filesystems.
1164 ASSERT(ip
->i_d
.di_aformat
== 0);
1165 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1168 xfs_trans_ijoin(tp
, ip
, 0);
1169 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1171 switch (ip
->i_d
.di_format
) {
1172 case XFS_DINODE_FMT_DEV
:
1173 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
1175 case XFS_DINODE_FMT_UUID
:
1176 ip
->i_d
.di_forkoff
= roundup(sizeof(uuid_t
), 8) >> 3;
1178 case XFS_DINODE_FMT_LOCAL
:
1179 case XFS_DINODE_FMT_EXTENTS
:
1180 case XFS_DINODE_FMT_BTREE
:
1181 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
1182 if (!ip
->i_d
.di_forkoff
)
1183 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
1184 else if (mp
->m_flags
& XFS_MOUNT_ATTR2
)
1193 ASSERT(ip
->i_afp
== NULL
);
1194 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
1195 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
1197 xfs_defer_init(&dfops
, &firstblock
);
1198 switch (ip
->i_d
.di_format
) {
1199 case XFS_DINODE_FMT_LOCAL
:
1200 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &firstblock
, &dfops
,
1203 case XFS_DINODE_FMT_EXTENTS
:
1204 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &firstblock
,
1207 case XFS_DINODE_FMT_BTREE
:
1208 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &firstblock
, &dfops
,
1216 xfs_trans_log_inode(tp
, ip
, logflags
);
1219 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
1220 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
1221 bool log_sb
= false;
1223 spin_lock(&mp
->m_sb_lock
);
1224 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
1225 xfs_sb_version_addattr(&mp
->m_sb
);
1228 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
1229 xfs_sb_version_addattr2(&mp
->m_sb
);
1232 spin_unlock(&mp
->m_sb_lock
);
1237 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1240 error
= xfs_trans_commit(tp
);
1241 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1245 xfs_defer_cancel(&dfops
);
1247 xfs_trans_cancel(tp
);
1248 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1253 * Internal and external extent tree search functions.
1257 * Read in the extents to if_extents.
1258 * All inode fields are set up by caller, we just traverse the btree
1259 * and copy the records in. If the file system cannot contain unwritten
1260 * extents, the records are checked for no "state" flags.
1263 xfs_bmap_read_extents(
1264 xfs_trans_t
*tp
, /* transaction pointer */
1265 xfs_inode_t
*ip
, /* incore inode */
1266 int whichfork
) /* data or attr fork */
1268 struct xfs_btree_block
*block
; /* current btree block */
1269 xfs_fsblock_t bno
; /* block # of "block" */
1270 xfs_buf_t
*bp
; /* buffer for "block" */
1271 int error
; /* error return value */
1272 xfs_exntfmt_t exntf
; /* XFS_EXTFMT_NOSTATE, if checking */
1273 xfs_extnum_t i
, j
; /* index into the extents list */
1274 xfs_ifork_t
*ifp
; /* fork structure */
1275 int level
; /* btree level, for checking */
1276 xfs_mount_t
*mp
; /* file system mount structure */
1277 __be64
*pp
; /* pointer to block address */
1279 xfs_extnum_t room
; /* number of entries there's room for */
1282 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1283 exntf
= (whichfork
!= XFS_DATA_FORK
) ? XFS_EXTFMT_NOSTATE
:
1284 XFS_EXTFMT_INODE(ip
);
1285 block
= ifp
->if_broot
;
1287 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1289 level
= be16_to_cpu(block
->bb_level
);
1291 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
1292 bno
= be64_to_cpu(*pp
);
1295 * Go down the tree until leaf level is reached, following the first
1296 * pointer (leftmost) at each level.
1298 while (level
-- > 0) {
1299 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1300 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1303 block
= XFS_BUF_TO_BLOCK(bp
);
1306 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
1307 bno
= be64_to_cpu(*pp
);
1308 XFS_WANT_CORRUPTED_GOTO(mp
,
1309 XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
1310 xfs_trans_brelse(tp
, bp
);
1313 * Here with bp and block set to the leftmost leaf node in the tree.
1315 room
= xfs_iext_count(ifp
);
1318 * Loop over all leaf nodes. Copy information to the extent records.
1321 xfs_bmbt_rec_t
*frp
;
1322 xfs_fsblock_t nextbno
;
1323 xfs_extnum_t num_recs
;
1326 num_recs
= xfs_btree_get_numrecs(block
);
1327 if (unlikely(i
+ num_recs
> room
)) {
1328 ASSERT(i
+ num_recs
<= room
);
1329 xfs_warn(ip
->i_mount
,
1330 "corrupt dinode %Lu, (btree extents).",
1331 (unsigned long long) ip
->i_ino
);
1332 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1333 XFS_ERRLEVEL_LOW
, ip
->i_mount
, block
);
1337 * Read-ahead the next leaf block, if any.
1339 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
1340 if (nextbno
!= NULLFSBLOCK
)
1341 xfs_btree_reada_bufl(mp
, nextbno
, 1,
1344 * Copy records into the extent records.
1346 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
1348 for (j
= 0; j
< num_recs
; j
++, i
++, frp
++) {
1349 xfs_bmbt_rec_host_t
*trp
= xfs_iext_get_ext(ifp
, i
);
1350 trp
->l0
= be64_to_cpu(frp
->l0
);
1351 trp
->l1
= be64_to_cpu(frp
->l1
);
1353 if (exntf
== XFS_EXTFMT_NOSTATE
) {
1355 * Check all attribute bmap btree records and
1356 * any "older" data bmap btree records for a
1357 * set bit in the "extent flag" position.
1359 if (unlikely(xfs_check_nostate_extents(ifp
,
1360 start
, num_recs
))) {
1361 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1367 xfs_trans_brelse(tp
, bp
);
1370 * If we've reached the end, stop.
1372 if (bno
== NULLFSBLOCK
)
1374 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1375 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1378 block
= XFS_BUF_TO_BLOCK(bp
);
1380 if (i
!= XFS_IFORK_NEXTENTS(ip
, whichfork
))
1381 return -EFSCORRUPTED
;
1382 ASSERT(i
== xfs_iext_count(ifp
));
1383 XFS_BMAP_TRACE_EXLIST(ip
, i
, whichfork
);
1386 xfs_trans_brelse(tp
, bp
);
1387 return -EFSCORRUPTED
;
1391 * Returns the file-relative block number of the first unused block(s)
1392 * in the file with at least "len" logically contiguous blocks free.
1393 * This is the lowest-address hole if the file has holes, else the first block
1394 * past the end of file.
1395 * Return 0 if the file is currently local (in-inode).
1398 xfs_bmap_first_unused(
1399 xfs_trans_t
*tp
, /* transaction pointer */
1400 xfs_inode_t
*ip
, /* incore inode */
1401 xfs_extlen_t len
, /* size of hole to find */
1402 xfs_fileoff_t
*first_unused
, /* unused block */
1403 int whichfork
) /* data or attr fork */
1405 int error
; /* error return value */
1406 int idx
; /* extent record index */
1407 xfs_ifork_t
*ifp
; /* inode fork pointer */
1408 xfs_fileoff_t lastaddr
; /* last block number seen */
1409 xfs_fileoff_t lowest
; /* lowest useful block */
1410 xfs_fileoff_t max
; /* starting useful block */
1411 xfs_fileoff_t off
; /* offset for this block */
1412 xfs_extnum_t nextents
; /* number of extent entries */
1414 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
||
1415 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
||
1416 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
1417 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
1421 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1422 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
1423 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
1425 lowest
= *first_unused
;
1426 nextents
= xfs_iext_count(ifp
);
1427 for (idx
= 0, lastaddr
= 0, max
= lowest
; idx
< nextents
; idx
++) {
1428 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, idx
);
1429 off
= xfs_bmbt_get_startoff(ep
);
1431 * See if the hole before this extent will work.
1433 if (off
>= lowest
+ len
&& off
- max
>= len
) {
1434 *first_unused
= max
;
1437 lastaddr
= off
+ xfs_bmbt_get_blockcount(ep
);
1438 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
1440 *first_unused
= max
;
1445 * Returns the file-relative block number of the last block - 1 before
1446 * last_block (input value) in the file.
1447 * This is not based on i_size, it is based on the extent records.
1448 * Returns 0 for local files, as they do not have extent records.
1451 xfs_bmap_last_before(
1452 struct xfs_trans
*tp
, /* transaction pointer */
1453 struct xfs_inode
*ip
, /* incore inode */
1454 xfs_fileoff_t
*last_block
, /* last block */
1455 int whichfork
) /* data or attr fork */
1457 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1458 struct xfs_bmbt_irec got
;
1462 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
1463 case XFS_DINODE_FMT_LOCAL
:
1466 case XFS_DINODE_FMT_BTREE
:
1467 case XFS_DINODE_FMT_EXTENTS
:
1473 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1474 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1479 if (xfs_iext_lookup_extent(ip
, ifp
, *last_block
- 1, &idx
, &got
)) {
1480 if (got
.br_startoff
<= *last_block
- 1)
1484 if (xfs_iext_get_extent(ifp
, idx
- 1, &got
)) {
1485 *last_block
= got
.br_startoff
+ got
.br_blockcount
;
1494 xfs_bmap_last_extent(
1495 struct xfs_trans
*tp
,
1496 struct xfs_inode
*ip
,
1498 struct xfs_bmbt_irec
*rec
,
1501 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1505 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1506 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1511 nextents
= xfs_iext_count(ifp
);
1512 if (nextents
== 0) {
1517 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, nextents
- 1), rec
);
1523 * Check the last inode extent to determine whether this allocation will result
1524 * in blocks being allocated at the end of the file. When we allocate new data
1525 * blocks at the end of the file which do not start at the previous data block,
1526 * we will try to align the new blocks at stripe unit boundaries.
1528 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1529 * at, or past the EOF.
1533 struct xfs_bmalloca
*bma
,
1536 struct xfs_bmbt_irec rec
;
1541 error
= xfs_bmap_last_extent(NULL
, bma
->ip
, whichfork
, &rec
,
1552 * Check if we are allocation or past the last extent, or at least into
1553 * the last delayed allocated extent.
1555 bma
->aeof
= bma
->offset
>= rec
.br_startoff
+ rec
.br_blockcount
||
1556 (bma
->offset
>= rec
.br_startoff
&&
1557 isnullstartblock(rec
.br_startblock
));
1562 * Returns the file-relative block number of the first block past eof in
1563 * the file. This is not based on i_size, it is based on the extent records.
1564 * Returns 0 for local files, as they do not have extent records.
1567 xfs_bmap_last_offset(
1568 struct xfs_inode
*ip
,
1569 xfs_fileoff_t
*last_block
,
1572 struct xfs_bmbt_irec rec
;
1578 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
)
1581 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
1582 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1585 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, &is_empty
);
1586 if (error
|| is_empty
)
1589 *last_block
= rec
.br_startoff
+ rec
.br_blockcount
;
1594 * Returns whether the selected fork of the inode has exactly one
1595 * block or not. For the data fork we check this matches di_size,
1596 * implying the file's range is 0..bsize-1.
1598 int /* 1=>1 block, 0=>otherwise */
1600 xfs_inode_t
*ip
, /* incore inode */
1601 int whichfork
) /* data or attr fork */
1603 xfs_bmbt_rec_host_t
*ep
; /* ptr to fork's extent */
1604 xfs_ifork_t
*ifp
; /* inode fork pointer */
1605 int rval
; /* return value */
1606 xfs_bmbt_irec_t s
; /* internal version of extent */
1609 if (whichfork
== XFS_DATA_FORK
)
1610 return XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
;
1612 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) != 1)
1614 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1616 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1617 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
1618 ep
= xfs_iext_get_ext(ifp
, 0);
1619 xfs_bmbt_get_all(ep
, &s
);
1620 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
1621 if (rval
&& whichfork
== XFS_DATA_FORK
)
1622 ASSERT(XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
);
1627 * Extent tree manipulation functions used during allocation.
1631 * Convert a delayed allocation to a real allocation.
1633 STATIC
int /* error */
1634 xfs_bmap_add_extent_delay_real(
1635 struct xfs_bmalloca
*bma
,
1638 struct xfs_bmbt_irec
*new = &bma
->got
;
1639 int diff
; /* temp value */
1640 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
1641 int error
; /* error return value */
1642 int i
; /* temp state */
1643 xfs_ifork_t
*ifp
; /* inode fork pointer */
1644 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1645 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1646 /* left is 0, right is 1, prev is 2 */
1647 int rval
=0; /* return value (logging flags) */
1648 int state
= 0;/* state bits, accessed thru macros */
1649 xfs_filblks_t da_new
; /* new count del alloc blocks used */
1650 xfs_filblks_t da_old
; /* old count del alloc blocks used */
1651 xfs_filblks_t temp
=0; /* value for da_new calculations */
1652 xfs_filblks_t temp2
=0;/* value for da_new calculations */
1653 int tmp_rval
; /* partial logging flags */
1654 struct xfs_mount
*mp
;
1655 xfs_extnum_t
*nextents
;
1657 mp
= bma
->ip
->i_mount
;
1658 ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
1659 ASSERT(whichfork
!= XFS_ATTR_FORK
);
1660 nextents
= (whichfork
== XFS_COW_FORK
? &bma
->ip
->i_cnextents
:
1661 &bma
->ip
->i_d
.di_nextents
);
1663 ASSERT(bma
->idx
>= 0);
1664 ASSERT(bma
->idx
<= xfs_iext_count(ifp
));
1665 ASSERT(!isnullstartblock(new->br_startblock
));
1667 (bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
1669 XFS_STATS_INC(mp
, xs_add_exlist
);
1675 if (whichfork
== XFS_COW_FORK
)
1676 state
|= BMAP_COWFORK
;
1679 * Set up a bunch of variables to make the tests simpler.
1681 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
1682 xfs_bmbt_get_all(ep
, &PREV
);
1683 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1684 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1685 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1687 da_old
= startblockval(PREV
.br_startblock
);
1691 * Set flags determining what part of the previous delayed allocation
1692 * extent is being replaced by a real allocation.
1694 if (PREV
.br_startoff
== new->br_startoff
)
1695 state
|= BMAP_LEFT_FILLING
;
1696 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1697 state
|= BMAP_RIGHT_FILLING
;
1700 * Check and set flags if this segment has a left neighbor.
1701 * Don't set contiguous if the combined extent would be too large.
1704 state
|= BMAP_LEFT_VALID
;
1705 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1), &LEFT
);
1707 if (isnullstartblock(LEFT
.br_startblock
))
1708 state
|= BMAP_LEFT_DELAY
;
1711 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1712 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1713 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1714 LEFT
.br_state
== new->br_state
&&
1715 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1716 state
|= BMAP_LEFT_CONTIG
;
1719 * Check and set flags if this segment has a right neighbor.
1720 * Don't set contiguous if the combined extent would be too large.
1721 * Also check for all-three-contiguous being too large.
1723 if (bma
->idx
< xfs_iext_count(ifp
) - 1) {
1724 state
|= BMAP_RIGHT_VALID
;
1725 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
+ 1), &RIGHT
);
1727 if (isnullstartblock(RIGHT
.br_startblock
))
1728 state
|= BMAP_RIGHT_DELAY
;
1731 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1732 new_endoff
== RIGHT
.br_startoff
&&
1733 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1734 new->br_state
== RIGHT
.br_state
&&
1735 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1736 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1737 BMAP_RIGHT_FILLING
)) !=
1738 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1739 BMAP_RIGHT_FILLING
) ||
1740 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1742 state
|= BMAP_RIGHT_CONTIG
;
1746 * Switch out based on the FILLING and CONTIG state bits.
1748 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1749 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1750 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1751 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1753 * Filling in all of a previously delayed allocation extent.
1754 * The left and right neighbors are both contiguous with new.
1757 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1758 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1759 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
1760 RIGHT
.br_blockcount
);
1761 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1763 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 2, state
);
1765 if (bma
->cur
== NULL
)
1766 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1768 rval
= XFS_ILOG_CORE
;
1769 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1770 RIGHT
.br_startblock
,
1771 RIGHT
.br_blockcount
, &i
);
1774 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1775 error
= xfs_btree_delete(bma
->cur
, &i
);
1778 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1779 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
1782 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1783 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1785 LEFT
.br_blockcount
+
1786 PREV
.br_blockcount
+
1787 RIGHT
.br_blockcount
, LEFT
.br_state
);
1793 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1795 * Filling in all of a previously delayed allocation extent.
1796 * The left neighbor is contiguous, the right is not.
1800 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1801 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1802 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
1803 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1805 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
1806 if (bma
->cur
== NULL
)
1807 rval
= XFS_ILOG_DEXT
;
1810 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
1811 LEFT
.br_startblock
, LEFT
.br_blockcount
,
1815 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1816 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1818 LEFT
.br_blockcount
+
1819 PREV
.br_blockcount
, LEFT
.br_state
);
1825 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1827 * Filling in all of a previously delayed allocation extent.
1828 * The right neighbor is contiguous, the left is not.
1830 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1831 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
1832 xfs_bmbt_set_blockcount(ep
,
1833 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
1834 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1836 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
1837 if (bma
->cur
== NULL
)
1838 rval
= XFS_ILOG_DEXT
;
1841 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1842 RIGHT
.br_startblock
,
1843 RIGHT
.br_blockcount
, &i
);
1846 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1847 error
= xfs_bmbt_update(bma
->cur
, PREV
.br_startoff
,
1849 PREV
.br_blockcount
+
1850 RIGHT
.br_blockcount
, PREV
.br_state
);
1856 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1858 * Filling in all of a previously delayed allocation extent.
1859 * Neither the left nor right neighbors are contiguous with
1862 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1863 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
1864 xfs_bmbt_set_state(ep
, new->br_state
);
1865 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1868 if (bma
->cur
== NULL
)
1869 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1871 rval
= XFS_ILOG_CORE
;
1872 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
1873 new->br_startblock
, new->br_blockcount
,
1877 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1878 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1879 error
= xfs_btree_insert(bma
->cur
, &i
);
1882 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1886 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1888 * Filling in the first part of a previous delayed allocation.
1889 * The left neighbor is contiguous.
1891 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
1892 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
1893 LEFT
.br_blockcount
+ new->br_blockcount
);
1894 xfs_bmbt_set_startoff(ep
,
1895 PREV
.br_startoff
+ new->br_blockcount
);
1896 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
1898 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1899 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1900 xfs_bmbt_set_blockcount(ep
, temp
);
1901 if (bma
->cur
== NULL
)
1902 rval
= XFS_ILOG_DEXT
;
1905 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
1906 LEFT
.br_startblock
, LEFT
.br_blockcount
,
1910 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1911 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1913 LEFT
.br_blockcount
+
1919 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1920 startblockval(PREV
.br_startblock
));
1921 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
1922 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1927 case BMAP_LEFT_FILLING
:
1929 * Filling in the first part of a previous delayed allocation.
1930 * The left neighbor is not contiguous.
1932 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1933 xfs_bmbt_set_startoff(ep
, new_endoff
);
1934 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1935 xfs_bmbt_set_blockcount(ep
, temp
);
1936 xfs_iext_insert(bma
->ip
, bma
->idx
, 1, new, state
);
1938 if (bma
->cur
== NULL
)
1939 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1941 rval
= XFS_ILOG_CORE
;
1942 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
1943 new->br_startblock
, new->br_blockcount
,
1947 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1948 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1949 error
= xfs_btree_insert(bma
->cur
, &i
);
1952 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1955 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1956 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1957 bma
->firstblock
, bma
->dfops
,
1958 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1963 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1964 startblockval(PREV
.br_startblock
) -
1965 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
1966 ep
= xfs_iext_get_ext(ifp
, bma
->idx
+ 1);
1967 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
1968 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
1971 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1973 * Filling in the last part of a previous delayed allocation.
1974 * The right neighbor is contiguous with the new allocation.
1976 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1977 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
1978 xfs_bmbt_set_blockcount(ep
, temp
);
1979 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, bma
->idx
+ 1),
1980 new->br_startoff
, new->br_startblock
,
1981 new->br_blockcount
+ RIGHT
.br_blockcount
,
1983 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
1984 if (bma
->cur
== NULL
)
1985 rval
= XFS_ILOG_DEXT
;
1988 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1989 RIGHT
.br_startblock
,
1990 RIGHT
.br_blockcount
, &i
);
1993 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1994 error
= xfs_bmbt_update(bma
->cur
, new->br_startoff
,
1996 new->br_blockcount
+
1997 RIGHT
.br_blockcount
,
2003 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
2004 startblockval(PREV
.br_startblock
));
2005 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2006 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
2007 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2012 case BMAP_RIGHT_FILLING
:
2014 * Filling in the last part of a previous delayed allocation.
2015 * The right neighbor is not contiguous.
2017 temp
= PREV
.br_blockcount
- new->br_blockcount
;
2018 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2019 xfs_bmbt_set_blockcount(ep
, temp
);
2020 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 1, new, state
);
2022 if (bma
->cur
== NULL
)
2023 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2025 rval
= XFS_ILOG_CORE
;
2026 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
2027 new->br_startblock
, new->br_blockcount
,
2031 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2032 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2033 error
= xfs_btree_insert(bma
->cur
, &i
);
2036 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2039 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
2040 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2041 bma
->firstblock
, bma
->dfops
, &bma
->cur
, 1,
2042 &tmp_rval
, whichfork
);
2047 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
2048 startblockval(PREV
.br_startblock
) -
2049 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
2050 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
2051 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
2052 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2059 * Filling in the middle part of a previous delayed allocation.
2060 * Contiguity is impossible here.
2061 * This case is avoided almost all the time.
2063 * We start with a delayed allocation:
2065 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2068 * and we are allocating:
2069 * +rrrrrrrrrrrrrrrrr+
2072 * and we set it up for insertion as:
2073 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2075 * PREV @ idx LEFT RIGHT
2076 * inserted at idx + 1
2078 temp
= new->br_startoff
- PREV
.br_startoff
;
2079 temp2
= PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
2080 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, 0, _THIS_IP_
);
2081 xfs_bmbt_set_blockcount(ep
, temp
); /* truncate PREV */
2083 RIGHT
.br_state
= PREV
.br_state
;
2084 RIGHT
.br_startblock
= nullstartblock(
2085 (int)xfs_bmap_worst_indlen(bma
->ip
, temp2
));
2086 RIGHT
.br_startoff
= new_endoff
;
2087 RIGHT
.br_blockcount
= temp2
;
2088 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2089 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 2, &LEFT
, state
);
2091 if (bma
->cur
== NULL
)
2092 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2094 rval
= XFS_ILOG_CORE
;
2095 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
2096 new->br_startblock
, new->br_blockcount
,
2100 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2101 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2102 error
= xfs_btree_insert(bma
->cur
, &i
);
2105 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2108 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
2109 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2110 bma
->firstblock
, bma
->dfops
, &bma
->cur
,
2111 1, &tmp_rval
, whichfork
);
2116 temp
= xfs_bmap_worst_indlen(bma
->ip
, temp
);
2117 temp2
= xfs_bmap_worst_indlen(bma
->ip
, temp2
);
2118 diff
= (int)(temp
+ temp2
- startblockval(PREV
.br_startblock
) -
2119 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
2121 error
= xfs_mod_fdblocks(bma
->ip
->i_mount
,
2122 -((int64_t)diff
), false);
2128 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
2129 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2130 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2131 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
2132 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, bma
->idx
+ 2),
2133 nullstartblock((int)temp2
));
2134 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
2137 da_new
= temp
+ temp2
;
2140 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2141 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2142 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2143 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2144 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2145 case BMAP_LEFT_CONTIG
:
2146 case BMAP_RIGHT_CONTIG
:
2148 * These cases are all impossible.
2153 /* add reverse mapping */
2154 error
= xfs_rmap_map_extent(mp
, bma
->dfops
, bma
->ip
, whichfork
, new);
2158 /* convert to a btree if necessary */
2159 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
2160 int tmp_logflags
; /* partial log flag return val */
2162 ASSERT(bma
->cur
== NULL
);
2163 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2164 bma
->firstblock
, bma
->dfops
, &bma
->cur
,
2165 da_old
> 0, &tmp_logflags
, whichfork
);
2166 bma
->logflags
|= tmp_logflags
;
2171 /* adjust for changes in reserved delayed indirect blocks */
2172 if (da_old
|| da_new
) {
2175 temp
+= bma
->cur
->bc_private
.b
.allocated
;
2176 ASSERT(temp
<= da_old
);
2178 xfs_mod_fdblocks(bma
->ip
->i_mount
,
2179 (int64_t)(da_old
- temp
), false);
2182 /* clear out the allocated field, done with it now in any case. */
2184 bma
->cur
->bc_private
.b
.allocated
= 0;
2186 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
2188 if (whichfork
!= XFS_COW_FORK
)
2189 bma
->logflags
|= rval
;
2197 * Convert an unwritten allocation to a real allocation or vice versa.
2199 STATIC
int /* error */
2200 xfs_bmap_add_extent_unwritten_real(
2201 struct xfs_trans
*tp
,
2202 xfs_inode_t
*ip
, /* incore inode pointer */
2204 xfs_extnum_t
*idx
, /* extent number to update/insert */
2205 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
2206 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
2207 xfs_fsblock_t
*first
, /* pointer to firstblock variable */
2208 struct xfs_defer_ops
*dfops
, /* list of extents to be freed */
2209 int *logflagsp
) /* inode logging flags */
2211 xfs_btree_cur_t
*cur
; /* btree cursor */
2212 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
2213 int error
; /* error return value */
2214 int i
; /* temp state */
2215 xfs_ifork_t
*ifp
; /* inode fork pointer */
2216 xfs_fileoff_t new_endoff
; /* end offset of new entry */
2217 xfs_exntst_t newext
; /* new extent state */
2218 xfs_exntst_t oldext
; /* old extent state */
2219 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
2220 /* left is 0, right is 1, prev is 2 */
2221 int rval
=0; /* return value (logging flags) */
2222 int state
= 0;/* state bits, accessed thru macros */
2223 struct xfs_mount
*mp
= ip
->i_mount
;
2228 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2229 if (whichfork
== XFS_COW_FORK
)
2230 state
|= BMAP_COWFORK
;
2233 ASSERT(*idx
<= xfs_iext_count(ifp
));
2234 ASSERT(!isnullstartblock(new->br_startblock
));
2236 XFS_STATS_INC(mp
, xs_add_exlist
);
2243 * Set up a bunch of variables to make the tests simpler.
2246 ep
= xfs_iext_get_ext(ifp
, *idx
);
2247 xfs_bmbt_get_all(ep
, &PREV
);
2248 newext
= new->br_state
;
2249 oldext
= (newext
== XFS_EXT_UNWRITTEN
) ?
2250 XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
2251 ASSERT(PREV
.br_state
== oldext
);
2252 new_endoff
= new->br_startoff
+ new->br_blockcount
;
2253 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
2254 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
2257 * Set flags determining what part of the previous oldext allocation
2258 * extent is being replaced by a newext allocation.
2260 if (PREV
.br_startoff
== new->br_startoff
)
2261 state
|= BMAP_LEFT_FILLING
;
2262 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
2263 state
|= BMAP_RIGHT_FILLING
;
2266 * Check and set flags if this segment has a left neighbor.
2267 * Don't set contiguous if the combined extent would be too large.
2270 state
|= BMAP_LEFT_VALID
;
2271 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &LEFT
);
2273 if (isnullstartblock(LEFT
.br_startblock
))
2274 state
|= BMAP_LEFT_DELAY
;
2277 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2278 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
2279 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
2280 LEFT
.br_state
== newext
&&
2281 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2282 state
|= BMAP_LEFT_CONTIG
;
2285 * Check and set flags if this segment has a right neighbor.
2286 * Don't set contiguous if the combined extent would be too large.
2287 * Also check for all-three-contiguous being too large.
2289 if (*idx
< xfs_iext_count(ifp
) - 1) {
2290 state
|= BMAP_RIGHT_VALID
;
2291 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
+ 1), &RIGHT
);
2292 if (isnullstartblock(RIGHT
.br_startblock
))
2293 state
|= BMAP_RIGHT_DELAY
;
2296 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2297 new_endoff
== RIGHT
.br_startoff
&&
2298 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
2299 newext
== RIGHT
.br_state
&&
2300 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
2301 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2302 BMAP_RIGHT_FILLING
)) !=
2303 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2304 BMAP_RIGHT_FILLING
) ||
2305 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
2307 state
|= BMAP_RIGHT_CONTIG
;
2310 * Switch out based on the FILLING and CONTIG state bits.
2312 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2313 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
2314 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2315 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2317 * Setting all of a previous oldext extent to newext.
2318 * The left and right neighbors are both contiguous with new.
2322 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2323 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
2324 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
2325 RIGHT
.br_blockcount
);
2326 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2328 xfs_iext_remove(ip
, *idx
+ 1, 2, state
);
2329 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2330 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 2);
2332 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2334 rval
= XFS_ILOG_CORE
;
2335 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
2336 RIGHT
.br_startblock
,
2337 RIGHT
.br_blockcount
, &i
)))
2339 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2340 if ((error
= xfs_btree_delete(cur
, &i
)))
2342 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2343 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2345 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2346 if ((error
= xfs_btree_delete(cur
, &i
)))
2348 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2349 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2351 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2352 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2354 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
2355 RIGHT
.br_blockcount
, LEFT
.br_state
)))
2360 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2362 * Setting all of a previous oldext extent to newext.
2363 * The left neighbor is contiguous, the right is not.
2367 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2368 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
2369 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
2370 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2372 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2373 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2374 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2376 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2378 rval
= XFS_ILOG_CORE
;
2379 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2380 PREV
.br_startblock
, PREV
.br_blockcount
,
2383 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2384 if ((error
= xfs_btree_delete(cur
, &i
)))
2386 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2387 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2389 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2390 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2392 LEFT
.br_blockcount
+ PREV
.br_blockcount
,
2398 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2400 * Setting all of a previous oldext extent to newext.
2401 * The right neighbor is contiguous, the left is not.
2403 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2404 xfs_bmbt_set_blockcount(ep
,
2405 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
2406 xfs_bmbt_set_state(ep
, newext
);
2407 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2408 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2409 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2410 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2412 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2414 rval
= XFS_ILOG_CORE
;
2415 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
2416 RIGHT
.br_startblock
,
2417 RIGHT
.br_blockcount
, &i
)))
2419 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2420 if ((error
= xfs_btree_delete(cur
, &i
)))
2422 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2423 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2425 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2426 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2428 new->br_blockcount
+ RIGHT
.br_blockcount
,
2434 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
2436 * Setting all of a previous oldext extent to newext.
2437 * Neither the left nor right neighbors are contiguous with
2440 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2441 xfs_bmbt_set_state(ep
, newext
);
2442 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2445 rval
= XFS_ILOG_DEXT
;
2448 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2449 new->br_startblock
, new->br_blockcount
,
2452 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2453 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2454 new->br_startblock
, new->br_blockcount
,
2460 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
2462 * Setting the first part of a previous oldext extent to newext.
2463 * The left neighbor is contiguous.
2465 trace_xfs_bmap_pre_update(ip
, *idx
- 1, state
, _THIS_IP_
);
2466 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
- 1),
2467 LEFT
.br_blockcount
+ new->br_blockcount
);
2468 xfs_bmbt_set_startoff(ep
,
2469 PREV
.br_startoff
+ new->br_blockcount
);
2470 trace_xfs_bmap_post_update(ip
, *idx
- 1, state
, _THIS_IP_
);
2472 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2473 xfs_bmbt_set_startblock(ep
,
2474 new->br_startblock
+ new->br_blockcount
);
2475 xfs_bmbt_set_blockcount(ep
,
2476 PREV
.br_blockcount
- new->br_blockcount
);
2477 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2482 rval
= XFS_ILOG_DEXT
;
2485 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2486 PREV
.br_startblock
, PREV
.br_blockcount
,
2489 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2490 if ((error
= xfs_bmbt_update(cur
,
2491 PREV
.br_startoff
+ new->br_blockcount
,
2492 PREV
.br_startblock
+ new->br_blockcount
,
2493 PREV
.br_blockcount
- new->br_blockcount
,
2496 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2498 error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2500 LEFT
.br_blockcount
+ new->br_blockcount
,
2507 case BMAP_LEFT_FILLING
:
2509 * Setting the first part of a previous oldext extent to newext.
2510 * The left neighbor is not contiguous.
2512 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2513 ASSERT(ep
&& xfs_bmbt_get_state(ep
) == oldext
);
2514 xfs_bmbt_set_startoff(ep
, new_endoff
);
2515 xfs_bmbt_set_blockcount(ep
,
2516 PREV
.br_blockcount
- new->br_blockcount
);
2517 xfs_bmbt_set_startblock(ep
,
2518 new->br_startblock
+ new->br_blockcount
);
2519 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2521 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2522 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2523 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2525 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2527 rval
= XFS_ILOG_CORE
;
2528 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2529 PREV
.br_startblock
, PREV
.br_blockcount
,
2532 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2533 if ((error
= xfs_bmbt_update(cur
,
2534 PREV
.br_startoff
+ new->br_blockcount
,
2535 PREV
.br_startblock
+ new->br_blockcount
,
2536 PREV
.br_blockcount
- new->br_blockcount
,
2539 cur
->bc_rec
.b
= *new;
2540 if ((error
= xfs_btree_insert(cur
, &i
)))
2542 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2546 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2548 * Setting the last part of a previous oldext extent to newext.
2549 * The right neighbor is contiguous with the new allocation.
2551 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2552 xfs_bmbt_set_blockcount(ep
,
2553 PREV
.br_blockcount
- new->br_blockcount
);
2554 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2558 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2559 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
2560 new->br_startoff
, new->br_startblock
,
2561 new->br_blockcount
+ RIGHT
.br_blockcount
, newext
);
2562 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2565 rval
= XFS_ILOG_DEXT
;
2568 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2570 PREV
.br_blockcount
, &i
)))
2572 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2573 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
2575 PREV
.br_blockcount
- new->br_blockcount
,
2578 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
2580 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2582 new->br_blockcount
+ RIGHT
.br_blockcount
,
2588 case BMAP_RIGHT_FILLING
:
2590 * Setting the last part of a previous oldext extent to newext.
2591 * The right neighbor is not contiguous.
2593 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2594 xfs_bmbt_set_blockcount(ep
,
2595 PREV
.br_blockcount
- new->br_blockcount
);
2596 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2599 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2601 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2602 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2604 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2606 rval
= XFS_ILOG_CORE
;
2607 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2608 PREV
.br_startblock
, PREV
.br_blockcount
,
2611 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2612 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
2614 PREV
.br_blockcount
- new->br_blockcount
,
2617 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2618 new->br_startblock
, new->br_blockcount
,
2621 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2622 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2623 if ((error
= xfs_btree_insert(cur
, &i
)))
2625 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2631 * Setting the middle part of a previous oldext extent to
2632 * newext. Contiguity is impossible here.
2633 * One extent becomes three extents.
2635 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2636 xfs_bmbt_set_blockcount(ep
,
2637 new->br_startoff
- PREV
.br_startoff
);
2638 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2641 r
[1].br_startoff
= new_endoff
;
2642 r
[1].br_blockcount
=
2643 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
2644 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
2645 r
[1].br_state
= oldext
;
2648 xfs_iext_insert(ip
, *idx
, 2, &r
[0], state
);
2650 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2651 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 2);
2653 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2655 rval
= XFS_ILOG_CORE
;
2656 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2657 PREV
.br_startblock
, PREV
.br_blockcount
,
2660 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2661 /* new right extent - oldext */
2662 if ((error
= xfs_bmbt_update(cur
, r
[1].br_startoff
,
2663 r
[1].br_startblock
, r
[1].br_blockcount
,
2666 /* new left extent - oldext */
2667 cur
->bc_rec
.b
= PREV
;
2668 cur
->bc_rec
.b
.br_blockcount
=
2669 new->br_startoff
- PREV
.br_startoff
;
2670 if ((error
= xfs_btree_insert(cur
, &i
)))
2672 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2674 * Reset the cursor to the position of the new extent
2675 * we are about to insert as we can't trust it after
2676 * the previous insert.
2678 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2679 new->br_startblock
, new->br_blockcount
,
2682 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2683 /* new middle extent - newext */
2684 cur
->bc_rec
.b
.br_state
= new->br_state
;
2685 if ((error
= xfs_btree_insert(cur
, &i
)))
2687 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2691 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2692 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2693 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2694 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2695 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2696 case BMAP_LEFT_CONTIG
:
2697 case BMAP_RIGHT_CONTIG
:
2699 * These cases are all impossible.
2704 /* update reverse mappings */
2705 error
= xfs_rmap_convert_extent(mp
, dfops
, ip
, whichfork
, new);
2709 /* convert to a btree if necessary */
2710 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2711 int tmp_logflags
; /* partial log flag return val */
2713 ASSERT(cur
== NULL
);
2714 error
= xfs_bmap_extents_to_btree(tp
, ip
, first
, dfops
, &cur
,
2715 0, &tmp_logflags
, whichfork
);
2716 *logflagsp
|= tmp_logflags
;
2721 /* clear out the allocated field, done with it now in any case. */
2723 cur
->bc_private
.b
.allocated
= 0;
2727 xfs_bmap_check_leaf_extents(*curp
, ip
, whichfork
);
2737 * Convert a hole to a delayed allocation.
2740 xfs_bmap_add_extent_hole_delay(
2741 xfs_inode_t
*ip
, /* incore inode pointer */
2743 xfs_extnum_t
*idx
, /* extent number to update/insert */
2744 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
2746 xfs_ifork_t
*ifp
; /* inode fork pointer */
2747 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2748 xfs_filblks_t newlen
=0; /* new indirect size */
2749 xfs_filblks_t oldlen
=0; /* old indirect size */
2750 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2751 int state
; /* state bits, accessed thru macros */
2752 xfs_filblks_t temp
=0; /* temp for indirect calculations */
2754 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2756 if (whichfork
== XFS_COW_FORK
)
2757 state
|= BMAP_COWFORK
;
2758 ASSERT(isnullstartblock(new->br_startblock
));
2761 * Check and set flags if this segment has a left neighbor
2764 state
|= BMAP_LEFT_VALID
;
2765 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &left
);
2767 if (isnullstartblock(left
.br_startblock
))
2768 state
|= BMAP_LEFT_DELAY
;
2772 * Check and set flags if the current (right) segment exists.
2773 * If it doesn't exist, we're converting the hole at end-of-file.
2775 if (*idx
< xfs_iext_count(ifp
)) {
2776 state
|= BMAP_RIGHT_VALID
;
2777 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &right
);
2779 if (isnullstartblock(right
.br_startblock
))
2780 state
|= BMAP_RIGHT_DELAY
;
2784 * Set contiguity flags on the left and right neighbors.
2785 * Don't let extents get too large, even if the pieces are contiguous.
2787 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
2788 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2789 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2790 state
|= BMAP_LEFT_CONTIG
;
2792 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
2793 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2794 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2795 (!(state
& BMAP_LEFT_CONTIG
) ||
2796 (left
.br_blockcount
+ new->br_blockcount
+
2797 right
.br_blockcount
<= MAXEXTLEN
)))
2798 state
|= BMAP_RIGHT_CONTIG
;
2801 * Switch out based on the contiguity flags.
2803 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2804 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2806 * New allocation is contiguous with delayed allocations
2807 * on the left and on the right.
2808 * Merge all three into a single extent record.
2811 temp
= left
.br_blockcount
+ new->br_blockcount
+
2812 right
.br_blockcount
;
2814 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2815 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
2816 oldlen
= startblockval(left
.br_startblock
) +
2817 startblockval(new->br_startblock
) +
2818 startblockval(right
.br_startblock
);
2819 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
2820 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
2821 nullstartblock((int)newlen
));
2822 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2824 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2827 case BMAP_LEFT_CONTIG
:
2829 * New allocation is contiguous with a delayed allocation
2831 * Merge the new allocation with the left neighbor.
2834 temp
= left
.br_blockcount
+ new->br_blockcount
;
2836 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2837 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
2838 oldlen
= startblockval(left
.br_startblock
) +
2839 startblockval(new->br_startblock
);
2840 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
2841 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
2842 nullstartblock((int)newlen
));
2843 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2846 case BMAP_RIGHT_CONTIG
:
2848 * New allocation is contiguous with a delayed allocation
2850 * Merge the new allocation with the right neighbor.
2852 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2853 temp
= new->br_blockcount
+ right
.br_blockcount
;
2854 oldlen
= startblockval(new->br_startblock
) +
2855 startblockval(right
.br_startblock
);
2856 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
2857 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
2859 nullstartblock((int)newlen
), temp
, right
.br_state
);
2860 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2865 * New allocation is not contiguous with another
2866 * delayed allocation.
2867 * Insert a new entry.
2869 oldlen
= newlen
= 0;
2870 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2873 if (oldlen
!= newlen
) {
2874 ASSERT(oldlen
> newlen
);
2875 xfs_mod_fdblocks(ip
->i_mount
, (int64_t)(oldlen
- newlen
),
2878 * Nothing to do for disk quota accounting here.
2884 * Convert a hole to a real allocation.
2886 STATIC
int /* error */
2887 xfs_bmap_add_extent_hole_real(
2888 struct xfs_bmalloca
*bma
,
2891 struct xfs_bmbt_irec
*new = &bma
->got
;
2892 int error
; /* error return value */
2893 int i
; /* temp state */
2894 xfs_ifork_t
*ifp
; /* inode fork pointer */
2895 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2896 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2897 int rval
=0; /* return value (logging flags) */
2898 int state
; /* state bits, accessed thru macros */
2899 struct xfs_mount
*mp
;
2901 mp
= bma
->ip
->i_mount
;
2902 ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
2904 ASSERT(bma
->idx
>= 0);
2905 ASSERT(bma
->idx
<= xfs_iext_count(ifp
));
2906 ASSERT(!isnullstartblock(new->br_startblock
));
2908 !(bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
2909 ASSERT(whichfork
!= XFS_COW_FORK
);
2911 XFS_STATS_INC(mp
, xs_add_exlist
);
2914 if (whichfork
== XFS_ATTR_FORK
)
2915 state
|= BMAP_ATTRFORK
;
2918 * Check and set flags if this segment has a left neighbor.
2921 state
|= BMAP_LEFT_VALID
;
2922 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1), &left
);
2923 if (isnullstartblock(left
.br_startblock
))
2924 state
|= BMAP_LEFT_DELAY
;
2928 * Check and set flags if this segment has a current value.
2929 * Not true if we're inserting into the "hole" at eof.
2931 if (bma
->idx
< xfs_iext_count(ifp
)) {
2932 state
|= BMAP_RIGHT_VALID
;
2933 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &right
);
2934 if (isnullstartblock(right
.br_startblock
))
2935 state
|= BMAP_RIGHT_DELAY
;
2939 * We're inserting a real allocation between "left" and "right".
2940 * Set the contiguity flags. Don't let extents get too large.
2942 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2943 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2944 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
2945 left
.br_state
== new->br_state
&&
2946 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2947 state
|= BMAP_LEFT_CONTIG
;
2949 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2950 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2951 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
2952 new->br_state
== right
.br_state
&&
2953 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2954 (!(state
& BMAP_LEFT_CONTIG
) ||
2955 left
.br_blockcount
+ new->br_blockcount
+
2956 right
.br_blockcount
<= MAXEXTLEN
))
2957 state
|= BMAP_RIGHT_CONTIG
;
2961 * Select which case we're in here, and implement it.
2963 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2964 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2966 * New allocation is contiguous with real allocations on the
2967 * left and on the right.
2968 * Merge all three into a single extent record.
2971 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2972 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
2973 left
.br_blockcount
+ new->br_blockcount
+
2974 right
.br_blockcount
);
2975 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2977 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
2979 XFS_IFORK_NEXT_SET(bma
->ip
, whichfork
,
2980 XFS_IFORK_NEXTENTS(bma
->ip
, whichfork
) - 1);
2981 if (bma
->cur
== NULL
) {
2982 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2984 rval
= XFS_ILOG_CORE
;
2985 error
= xfs_bmbt_lookup_eq(bma
->cur
, right
.br_startoff
,
2986 right
.br_startblock
, right
.br_blockcount
,
2990 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2991 error
= xfs_btree_delete(bma
->cur
, &i
);
2994 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2995 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
2998 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2999 error
= xfs_bmbt_update(bma
->cur
, left
.br_startoff
,
3001 left
.br_blockcount
+
3002 new->br_blockcount
+
3003 right
.br_blockcount
,
3010 case BMAP_LEFT_CONTIG
:
3012 * New allocation is contiguous with a real allocation
3014 * Merge the new allocation with the left neighbor.
3017 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3018 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
3019 left
.br_blockcount
+ new->br_blockcount
);
3020 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3022 if (bma
->cur
== NULL
) {
3023 rval
= xfs_ilog_fext(whichfork
);
3026 error
= xfs_bmbt_lookup_eq(bma
->cur
, left
.br_startoff
,
3027 left
.br_startblock
, left
.br_blockcount
,
3031 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
3032 error
= xfs_bmbt_update(bma
->cur
, left
.br_startoff
,
3034 left
.br_blockcount
+
3042 case BMAP_RIGHT_CONTIG
:
3044 * New allocation is contiguous with a real allocation
3046 * Merge the new allocation with the right neighbor.
3048 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3049 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, bma
->idx
),
3050 new->br_startoff
, new->br_startblock
,
3051 new->br_blockcount
+ right
.br_blockcount
,
3053 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3055 if (bma
->cur
== NULL
) {
3056 rval
= xfs_ilog_fext(whichfork
);
3059 error
= xfs_bmbt_lookup_eq(bma
->cur
,
3061 right
.br_startblock
,
3062 right
.br_blockcount
, &i
);
3065 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
3066 error
= xfs_bmbt_update(bma
->cur
, new->br_startoff
,
3068 new->br_blockcount
+
3069 right
.br_blockcount
,
3078 * New allocation is not contiguous with another
3080 * Insert a new entry.
3082 xfs_iext_insert(bma
->ip
, bma
->idx
, 1, new, state
);
3083 XFS_IFORK_NEXT_SET(bma
->ip
, whichfork
,
3084 XFS_IFORK_NEXTENTS(bma
->ip
, whichfork
) + 1);
3085 if (bma
->cur
== NULL
) {
3086 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
3088 rval
= XFS_ILOG_CORE
;
3089 error
= xfs_bmbt_lookup_eq(bma
->cur
,
3092 new->br_blockcount
, &i
);
3095 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
3096 bma
->cur
->bc_rec
.b
.br_state
= new->br_state
;
3097 error
= xfs_btree_insert(bma
->cur
, &i
);
3100 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
3105 /* add reverse mapping */
3106 error
= xfs_rmap_map_extent(mp
, bma
->dfops
, bma
->ip
, whichfork
, new);
3110 /* convert to a btree if necessary */
3111 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
3112 int tmp_logflags
; /* partial log flag return val */
3114 ASSERT(bma
->cur
== NULL
);
3115 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
3116 bma
->firstblock
, bma
->dfops
, &bma
->cur
,
3117 0, &tmp_logflags
, whichfork
);
3118 bma
->logflags
|= tmp_logflags
;
3123 /* clear out the allocated field, done with it now in any case. */
3125 bma
->cur
->bc_private
.b
.allocated
= 0;
3127 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
3129 bma
->logflags
|= rval
;
3134 * Functions used in the extent read, allocate and remove paths
3138 * Adjust the size of the new extent based on di_extsize and rt extsize.
3141 xfs_bmap_extsize_align(
3143 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
3144 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
3145 xfs_extlen_t extsz
, /* align to this extent size */
3146 int rt
, /* is this a realtime inode? */
3147 int eof
, /* is extent at end-of-file? */
3148 int delay
, /* creating delalloc extent? */
3149 int convert
, /* overwriting unwritten extent? */
3150 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
3151 xfs_extlen_t
*lenp
) /* in/out: aligned length */
3153 xfs_fileoff_t orig_off
; /* original offset */
3154 xfs_extlen_t orig_alen
; /* original length */
3155 xfs_fileoff_t orig_end
; /* original off+len */
3156 xfs_fileoff_t nexto
; /* next file offset */
3157 xfs_fileoff_t prevo
; /* previous file offset */
3158 xfs_fileoff_t align_off
; /* temp for offset */
3159 xfs_extlen_t align_alen
; /* temp for length */
3160 xfs_extlen_t temp
; /* temp for calculations */
3165 orig_off
= align_off
= *offp
;
3166 orig_alen
= align_alen
= *lenp
;
3167 orig_end
= orig_off
+ orig_alen
;
3170 * If this request overlaps an existing extent, then don't
3171 * attempt to perform any additional alignment.
3173 if (!delay
&& !eof
&&
3174 (orig_off
>= gotp
->br_startoff
) &&
3175 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
3180 * If the file offset is unaligned vs. the extent size
3181 * we need to align it. This will be possible unless
3182 * the file was previously written with a kernel that didn't
3183 * perform this alignment, or if a truncate shot us in the
3186 temp
= do_mod(orig_off
, extsz
);
3192 /* Same adjustment for the end of the requested area. */
3193 temp
= (align_alen
% extsz
);
3195 align_alen
+= extsz
- temp
;
3198 * For large extent hint sizes, the aligned extent might be larger than
3199 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3200 * the length back under MAXEXTLEN. The outer allocation loops handle
3201 * short allocation just fine, so it is safe to do this. We only want to
3202 * do it when we are forced to, though, because it means more allocation
3203 * operations are required.
3205 while (align_alen
> MAXEXTLEN
)
3206 align_alen
-= extsz
;
3207 ASSERT(align_alen
<= MAXEXTLEN
);
3210 * If the previous block overlaps with this proposed allocation
3211 * then move the start forward without adjusting the length.
3213 if (prevp
->br_startoff
!= NULLFILEOFF
) {
3214 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
3215 prevo
= prevp
->br_startoff
;
3217 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
3220 if (align_off
!= orig_off
&& align_off
< prevo
)
3223 * If the next block overlaps with this proposed allocation
3224 * then move the start back without adjusting the length,
3225 * but not before offset 0.
3226 * This may of course make the start overlap previous block,
3227 * and if we hit the offset 0 limit then the next block
3228 * can still overlap too.
3230 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
3231 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
3232 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
3233 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
3235 nexto
= gotp
->br_startoff
;
3237 nexto
= NULLFILEOFF
;
3239 align_off
+ align_alen
!= orig_end
&&
3240 align_off
+ align_alen
> nexto
)
3241 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
3243 * If we're now overlapping the next or previous extent that
3244 * means we can't fit an extsz piece in this hole. Just move
3245 * the start forward to the first valid spot and set
3246 * the length so we hit the end.
3248 if (align_off
!= orig_off
&& align_off
< prevo
)
3250 if (align_off
+ align_alen
!= orig_end
&&
3251 align_off
+ align_alen
> nexto
&&
3252 nexto
!= NULLFILEOFF
) {
3253 ASSERT(nexto
> prevo
);
3254 align_alen
= nexto
- align_off
;
3258 * If realtime, and the result isn't a multiple of the realtime
3259 * extent size we need to remove blocks until it is.
3261 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
3263 * We're not covering the original request, or
3264 * we won't be able to once we fix the length.
3266 if (orig_off
< align_off
||
3267 orig_end
> align_off
+ align_alen
||
3268 align_alen
- temp
< orig_alen
)
3271 * Try to fix it by moving the start up.
3273 if (align_off
+ temp
<= orig_off
) {
3278 * Try to fix it by moving the end in.
3280 else if (align_off
+ align_alen
- temp
>= orig_end
)
3283 * Set the start to the minimum then trim the length.
3286 align_alen
-= orig_off
- align_off
;
3287 align_off
= orig_off
;
3288 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
3291 * Result doesn't cover the request, fail it.
3293 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
3296 ASSERT(orig_off
>= align_off
);
3297 /* see MAXEXTLEN handling above */
3298 ASSERT(orig_end
<= align_off
+ align_alen
||
3299 align_alen
+ extsz
> MAXEXTLEN
);
3303 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
3304 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
3305 if (prevp
->br_startoff
!= NULLFILEOFF
)
3306 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
3314 #define XFS_ALLOC_GAP_UNITS 4
3318 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3320 xfs_fsblock_t adjust
; /* adjustment to block numbers */
3321 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3322 xfs_mount_t
*mp
; /* mount point structure */
3323 int nullfb
; /* true if ap->firstblock isn't set */
3324 int rt
; /* true if inode is realtime */
3326 #define ISVALID(x,y) \
3328 (x) < mp->m_sb.sb_rblocks : \
3329 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3330 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3331 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3333 mp
= ap
->ip
->i_mount
;
3334 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
3335 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) &&
3336 xfs_alloc_is_userdata(ap
->datatype
);
3337 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
3339 * If allocating at eof, and there's a previous real block,
3340 * try to use its last block as our starting point.
3342 if (ap
->eof
&& ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3343 !isnullstartblock(ap
->prev
.br_startblock
) &&
3344 ISVALID(ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
,
3345 ap
->prev
.br_startblock
)) {
3346 ap
->blkno
= ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
;
3348 * Adjust for the gap between prevp and us.
3350 adjust
= ap
->offset
-
3351 (ap
->prev
.br_startoff
+ ap
->prev
.br_blockcount
);
3353 ISVALID(ap
->blkno
+ adjust
, ap
->prev
.br_startblock
))
3354 ap
->blkno
+= adjust
;
3357 * If not at eof, then compare the two neighbor blocks.
3358 * Figure out whether either one gives us a good starting point,
3359 * and pick the better one.
3361 else if (!ap
->eof
) {
3362 xfs_fsblock_t gotbno
; /* right side block number */
3363 xfs_fsblock_t gotdiff
=0; /* right side difference */
3364 xfs_fsblock_t prevbno
; /* left side block number */
3365 xfs_fsblock_t prevdiff
=0; /* left side difference */
3368 * If there's a previous (left) block, select a requested
3369 * start block based on it.
3371 if (ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3372 !isnullstartblock(ap
->prev
.br_startblock
) &&
3373 (prevbno
= ap
->prev
.br_startblock
+
3374 ap
->prev
.br_blockcount
) &&
3375 ISVALID(prevbno
, ap
->prev
.br_startblock
)) {
3377 * Calculate gap to end of previous block.
3379 adjust
= prevdiff
= ap
->offset
-
3380 (ap
->prev
.br_startoff
+
3381 ap
->prev
.br_blockcount
);
3383 * Figure the startblock based on the previous block's
3384 * end and the gap size.
3386 * If the gap is large relative to the piece we're
3387 * allocating, or using it gives us an invalid block
3388 * number, then just use the end of the previous block.
3390 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3391 ISVALID(prevbno
+ prevdiff
,
3392 ap
->prev
.br_startblock
))
3397 * If the firstblock forbids it, can't use it,
3400 if (!rt
&& !nullfb
&&
3401 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
3402 prevbno
= NULLFSBLOCK
;
3405 * No previous block or can't follow it, just default.
3408 prevbno
= NULLFSBLOCK
;
3410 * If there's a following (right) block, select a requested
3411 * start block based on it.
3413 if (!isnullstartblock(ap
->got
.br_startblock
)) {
3415 * Calculate gap to start of next block.
3417 adjust
= gotdiff
= ap
->got
.br_startoff
- ap
->offset
;
3419 * Figure the startblock based on the next block's
3420 * start and the gap size.
3422 gotbno
= ap
->got
.br_startblock
;
3425 * If the gap is large relative to the piece we're
3426 * allocating, or using it gives us an invalid block
3427 * number, then just use the start of the next block
3428 * offset by our length.
3430 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3431 ISVALID(gotbno
- gotdiff
, gotbno
))
3433 else if (ISVALID(gotbno
- ap
->length
, gotbno
)) {
3434 gotbno
-= ap
->length
;
3435 gotdiff
+= adjust
- ap
->length
;
3439 * If the firstblock forbids it, can't use it,
3442 if (!rt
&& !nullfb
&&
3443 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
3444 gotbno
= NULLFSBLOCK
;
3447 * No next block, just default.
3450 gotbno
= NULLFSBLOCK
;
3452 * If both valid, pick the better one, else the only good
3453 * one, else ap->blkno is already set (to 0 or the inode block).
3455 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
3456 ap
->blkno
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
3457 else if (prevbno
!= NULLFSBLOCK
)
3458 ap
->blkno
= prevbno
;
3459 else if (gotbno
!= NULLFSBLOCK
)
3466 xfs_bmap_longest_free_extent(
3467 struct xfs_trans
*tp
,
3472 struct xfs_mount
*mp
= tp
->t_mountp
;
3473 struct xfs_perag
*pag
;
3474 xfs_extlen_t longest
;
3477 pag
= xfs_perag_get(mp
, ag
);
3478 if (!pag
->pagf_init
) {
3479 error
= xfs_alloc_pagf_init(mp
, tp
, ag
, XFS_ALLOC_FLAG_TRYLOCK
);
3483 if (!pag
->pagf_init
) {
3489 longest
= xfs_alloc_longest_free_extent(mp
, pag
,
3490 xfs_alloc_min_freelist(mp
, pag
),
3491 xfs_ag_resv_needed(pag
, XFS_AG_RESV_NONE
));
3492 if (*blen
< longest
)
3501 xfs_bmap_select_minlen(
3502 struct xfs_bmalloca
*ap
,
3503 struct xfs_alloc_arg
*args
,
3507 if (notinit
|| *blen
< ap
->minlen
) {
3509 * Since we did a BUF_TRYLOCK above, it is possible that
3510 * there is space for this request.
3512 args
->minlen
= ap
->minlen
;
3513 } else if (*blen
< args
->maxlen
) {
3515 * If the best seen length is less than the request length,
3516 * use the best as the minimum.
3518 args
->minlen
= *blen
;
3521 * Otherwise we've seen an extent as big as maxlen, use that
3524 args
->minlen
= args
->maxlen
;
3529 xfs_bmap_btalloc_nullfb(
3530 struct xfs_bmalloca
*ap
,
3531 struct xfs_alloc_arg
*args
,
3534 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3535 xfs_agnumber_t ag
, startag
;
3539 args
->type
= XFS_ALLOCTYPE_START_BNO
;
3540 args
->total
= ap
->total
;
3542 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3543 if (startag
== NULLAGNUMBER
)
3546 while (*blen
< args
->maxlen
) {
3547 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3552 if (++ag
== mp
->m_sb
.sb_agcount
)
3558 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3563 xfs_bmap_btalloc_filestreams(
3564 struct xfs_bmalloca
*ap
,
3565 struct xfs_alloc_arg
*args
,
3568 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3573 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
3574 args
->total
= ap
->total
;
3576 ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3577 if (ag
== NULLAGNUMBER
)
3580 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
, ¬init
);
3584 if (*blen
< args
->maxlen
) {
3585 error
= xfs_filestream_new_ag(ap
, &ag
);
3589 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3596 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3599 * Set the failure fallback case to look in the selected AG as stream
3602 ap
->blkno
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3608 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3610 xfs_mount_t
*mp
; /* mount point structure */
3611 xfs_alloctype_t atype
= 0; /* type for allocation routines */
3612 xfs_extlen_t align
= 0; /* minimum allocation alignment */
3613 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3615 xfs_alloc_arg_t args
;
3617 xfs_extlen_t nextminlen
= 0;
3618 int nullfb
; /* true if ap->firstblock isn't set */
3626 mp
= ap
->ip
->i_mount
;
3628 /* stripe alignment for allocation is determined by mount parameters */
3630 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
3631 stripe_align
= mp
->m_swidth
;
3632 else if (mp
->m_dalign
)
3633 stripe_align
= mp
->m_dalign
;
3635 if (ap
->flags
& XFS_BMAPI_COWFORK
)
3636 align
= xfs_get_cowextsz_hint(ap
->ip
);
3637 else if (xfs_alloc_is_userdata(ap
->datatype
))
3638 align
= xfs_get_extsz_hint(ap
->ip
);
3640 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
3641 align
, 0, ap
->eof
, 0, ap
->conv
,
3642 &ap
->offset
, &ap
->length
);
3648 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
3649 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
3651 if (xfs_alloc_is_userdata(ap
->datatype
) &&
3652 xfs_inode_is_filestream(ap
->ip
)) {
3653 ag
= xfs_filestream_lookup_ag(ap
->ip
);
3654 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
3655 ap
->blkno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3657 ap
->blkno
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
3660 ap
->blkno
= *ap
->firstblock
;
3662 xfs_bmap_adjacent(ap
);
3665 * If allowed, use ap->blkno; otherwise must use firstblock since
3666 * it's in the right allocation group.
3668 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->blkno
) == fb_agno
)
3671 ap
->blkno
= *ap
->firstblock
;
3673 * Normal allocation, done through xfs_alloc_vextent.
3675 tryagain
= isaligned
= 0;
3676 memset(&args
, 0, sizeof(args
));
3679 args
.fsbno
= ap
->blkno
;
3680 xfs_rmap_skip_owner_update(&args
.oinfo
);
3682 /* Trim the allocation back to the maximum an AG can fit. */
3683 args
.maxlen
= MIN(ap
->length
, mp
->m_ag_max_usable
);
3684 args
.firstblock
= *ap
->firstblock
;
3688 * Search for an allocation group with a single extent large
3689 * enough for the request. If one isn't found, then adjust
3690 * the minimum allocation size to the largest space found.
3692 if (xfs_alloc_is_userdata(ap
->datatype
) &&
3693 xfs_inode_is_filestream(ap
->ip
))
3694 error
= xfs_bmap_btalloc_filestreams(ap
, &args
, &blen
);
3696 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
3699 } else if (ap
->dfops
->dop_low
) {
3700 if (xfs_inode_is_filestream(ap
->ip
))
3701 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3703 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3704 args
.total
= args
.minlen
= ap
->minlen
;
3706 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3707 args
.total
= ap
->total
;
3708 args
.minlen
= ap
->minlen
;
3710 /* apply extent size hints if obtained earlier */
3713 if ((args
.mod
= (xfs_extlen_t
)do_mod(ap
->offset
, args
.prod
)))
3714 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
3715 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_SIZE
) {
3719 args
.prod
= PAGE_SIZE
>> mp
->m_sb
.sb_blocklog
;
3720 if ((args
.mod
= (xfs_extlen_t
)(do_mod(ap
->offset
, args
.prod
))))
3721 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
3724 * If we are not low on available data blocks, and the
3725 * underlying logical volume manager is a stripe, and
3726 * the file offset is zero then try to allocate data
3727 * blocks on stripe unit boundary.
3728 * NOTE: ap->aeof is only set if the allocation length
3729 * is >= the stripe unit and the allocation offset is
3730 * at the end of file.
3732 if (!ap
->dfops
->dop_low
&& ap
->aeof
) {
3734 args
.alignment
= stripe_align
;
3738 * Adjust for alignment
3740 if (blen
> args
.alignment
&& blen
<= args
.maxlen
)
3741 args
.minlen
= blen
- args
.alignment
;
3742 args
.minalignslop
= 0;
3745 * First try an exact bno allocation.
3746 * If it fails then do a near or start bno
3747 * allocation with alignment turned on.
3751 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
3754 * Compute the minlen+alignment for the
3755 * next case. Set slop so that the value
3756 * of minlen+alignment+slop doesn't go up
3757 * between the calls.
3759 if (blen
> stripe_align
&& blen
<= args
.maxlen
)
3760 nextminlen
= blen
- stripe_align
;
3762 nextminlen
= args
.minlen
;
3763 if (nextminlen
+ stripe_align
> args
.minlen
+ 1)
3765 nextminlen
+ stripe_align
-
3768 args
.minalignslop
= 0;
3772 args
.minalignslop
= 0;
3774 args
.minleft
= ap
->minleft
;
3775 args
.wasdel
= ap
->wasdel
;
3776 args
.resv
= XFS_AG_RESV_NONE
;
3777 args
.datatype
= ap
->datatype
;
3778 if (ap
->datatype
& XFS_ALLOC_USERDATA_ZERO
)
3781 error
= xfs_alloc_vextent(&args
);
3785 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
3787 * Exact allocation failed. Now try with alignment
3791 args
.fsbno
= ap
->blkno
;
3792 args
.alignment
= stripe_align
;
3793 args
.minlen
= nextminlen
;
3794 args
.minalignslop
= 0;
3796 if ((error
= xfs_alloc_vextent(&args
)))
3799 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
3801 * allocation failed, so turn off alignment and
3805 args
.fsbno
= ap
->blkno
;
3807 if ((error
= xfs_alloc_vextent(&args
)))
3810 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
3811 args
.minlen
> ap
->minlen
) {
3812 args
.minlen
= ap
->minlen
;
3813 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3814 args
.fsbno
= ap
->blkno
;
3815 if ((error
= xfs_alloc_vextent(&args
)))
3818 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
3820 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3821 args
.total
= ap
->minlen
;
3822 if ((error
= xfs_alloc_vextent(&args
)))
3824 ap
->dfops
->dop_low
= true;
3826 if (args
.fsbno
!= NULLFSBLOCK
) {
3828 * check the allocation happened at the same or higher AG than
3829 * the first block that was allocated.
3831 ASSERT(*ap
->firstblock
== NULLFSBLOCK
||
3832 XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
) ==
3833 XFS_FSB_TO_AGNO(mp
, args
.fsbno
) ||
3834 (ap
->dfops
->dop_low
&&
3835 XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
) <
3836 XFS_FSB_TO_AGNO(mp
, args
.fsbno
)));
3838 ap
->blkno
= args
.fsbno
;
3839 if (*ap
->firstblock
== NULLFSBLOCK
)
3840 *ap
->firstblock
= args
.fsbno
;
3841 ASSERT(nullfb
|| fb_agno
== args
.agno
||
3842 (ap
->dfops
->dop_low
&& fb_agno
< args
.agno
));
3843 ap
->length
= args
.len
;
3844 if (!(ap
->flags
& XFS_BMAPI_COWFORK
))
3845 ap
->ip
->i_d
.di_nblocks
+= args
.len
;
3846 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
3848 ap
->ip
->i_delayed_blks
-= args
.len
;
3850 * Adjust the disk quota also. This was reserved
3853 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
3854 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
:
3855 XFS_TRANS_DQ_BCOUNT
,
3858 ap
->blkno
= NULLFSBLOCK
;
3865 * For a remap operation, just "allocate" an extent at the address that the
3866 * caller passed in, and ensure that the AGFL is the right size. The caller
3867 * will then map the "allocated" extent into the file somewhere.
3870 xfs_bmap_remap_alloc(
3871 struct xfs_bmalloca
*ap
)
3873 struct xfs_trans
*tp
= ap
->tp
;
3874 struct xfs_mount
*mp
= tp
->t_mountp
;
3876 struct xfs_alloc_arg args
;
3880 * validate that the block number is legal - the enables us to detect
3881 * and handle a silent filesystem corruption rather than crashing.
3883 memset(&args
, 0, sizeof(struct xfs_alloc_arg
));
3885 args
.mp
= ap
->tp
->t_mountp
;
3886 bno
= *ap
->firstblock
;
3887 args
.agno
= XFS_FSB_TO_AGNO(mp
, bno
);
3888 args
.agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
3889 if (args
.agno
>= mp
->m_sb
.sb_agcount
||
3890 args
.agbno
>= mp
->m_sb
.sb_agblocks
)
3891 return -EFSCORRUPTED
;
3893 /* "Allocate" the extent from the range we passed in. */
3894 trace_xfs_bmap_remap_alloc(ap
->ip
, *ap
->firstblock
, ap
->length
);
3896 ap
->ip
->i_d
.di_nblocks
+= ap
->length
;
3897 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
3899 /* Fix the freelist, like a real allocator does. */
3900 args
.datatype
= ap
->datatype
;
3901 args
.pag
= xfs_perag_get(args
.mp
, args
.agno
);
3905 * The freelist fixing code will decline the allocation if
3906 * the size and shape of the free space doesn't allow for
3907 * allocating the extent and updating all the metadata that
3908 * happens during an allocation. We're remapping, not
3909 * allocating, so skip that check by pretending to be freeing.
3911 error
= xfs_alloc_fix_freelist(&args
, XFS_ALLOC_FLAG_FREEING
);
3912 xfs_perag_put(args
.pag
);
3914 trace_xfs_bmap_remap_alloc_error(ap
->ip
, error
, _RET_IP_
);
3919 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3920 * It figures out where to ask the underlying allocator to put the new extent.
3924 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3926 if (ap
->flags
& XFS_BMAPI_REMAP
)
3927 return xfs_bmap_remap_alloc(ap
);
3928 if (XFS_IS_REALTIME_INODE(ap
->ip
) &&
3929 xfs_alloc_is_userdata(ap
->datatype
))
3930 return xfs_bmap_rtalloc(ap
);
3931 return xfs_bmap_btalloc(ap
);
3934 /* Trim extent to fit a logical block range. */
3937 struct xfs_bmbt_irec
*irec
,
3941 xfs_fileoff_t distance
;
3942 xfs_fileoff_t end
= bno
+ len
;
3944 if (irec
->br_startoff
+ irec
->br_blockcount
<= bno
||
3945 irec
->br_startoff
>= end
) {
3946 irec
->br_blockcount
= 0;
3950 if (irec
->br_startoff
< bno
) {
3951 distance
= bno
- irec
->br_startoff
;
3952 if (isnullstartblock(irec
->br_startblock
))
3953 irec
->br_startblock
= DELAYSTARTBLOCK
;
3954 if (irec
->br_startblock
!= DELAYSTARTBLOCK
&&
3955 irec
->br_startblock
!= HOLESTARTBLOCK
)
3956 irec
->br_startblock
+= distance
;
3957 irec
->br_startoff
+= distance
;
3958 irec
->br_blockcount
-= distance
;
3961 if (end
< irec
->br_startoff
+ irec
->br_blockcount
) {
3962 distance
= irec
->br_startoff
+ irec
->br_blockcount
- end
;
3963 irec
->br_blockcount
-= distance
;
3968 * Trim the returned map to the required bounds
3972 struct xfs_bmbt_irec
*mval
,
3973 struct xfs_bmbt_irec
*got
,
3981 if ((flags
& XFS_BMAPI_ENTIRE
) ||
3982 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
3984 if (isnullstartblock(got
->br_startblock
))
3985 mval
->br_startblock
= DELAYSTARTBLOCK
;
3991 ASSERT((*bno
>= obno
) || (n
== 0));
3993 mval
->br_startoff
= *bno
;
3994 if (isnullstartblock(got
->br_startblock
))
3995 mval
->br_startblock
= DELAYSTARTBLOCK
;
3997 mval
->br_startblock
= got
->br_startblock
+
3998 (*bno
- got
->br_startoff
);
4000 * Return the minimum of what we got and what we asked for for
4001 * the length. We can use the len variable here because it is
4002 * modified below and we could have been there before coming
4003 * here if the first part of the allocation didn't overlap what
4006 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
4007 got
->br_blockcount
- (*bno
- got
->br_startoff
));
4008 mval
->br_state
= got
->br_state
;
4009 ASSERT(mval
->br_blockcount
<= len
);
4014 * Update and validate the extent map to return
4017 xfs_bmapi_update_map(
4018 struct xfs_bmbt_irec
**map
,
4026 xfs_bmbt_irec_t
*mval
= *map
;
4028 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
4029 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
4030 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
4031 (mval
->br_startoff
< obno
));
4033 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
4035 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
4036 /* update previous map with new information */
4037 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
4038 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
4039 ASSERT(mval
->br_state
== mval
[-1].br_state
);
4040 mval
[-1].br_blockcount
= mval
->br_blockcount
;
4041 mval
[-1].br_state
= mval
->br_state
;
4042 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
4043 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
4044 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
4045 mval
->br_startblock
== mval
[-1].br_startblock
+
4046 mval
[-1].br_blockcount
&&
4047 ((flags
& XFS_BMAPI_IGSTATE
) ||
4048 mval
[-1].br_state
== mval
->br_state
)) {
4049 ASSERT(mval
->br_startoff
==
4050 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
4051 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4052 } else if (*n
> 0 &&
4053 mval
->br_startblock
== DELAYSTARTBLOCK
&&
4054 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
4055 mval
->br_startoff
==
4056 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
4057 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4058 mval
[-1].br_state
= mval
->br_state
;
4059 } else if (!((*n
== 0) &&
4060 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
4069 * Map file blocks to filesystem blocks without allocation.
4073 struct xfs_inode
*ip
,
4076 struct xfs_bmbt_irec
*mval
,
4080 struct xfs_mount
*mp
= ip
->i_mount
;
4081 struct xfs_ifork
*ifp
;
4082 struct xfs_bmbt_irec got
;
4089 int whichfork
= xfs_bmapi_whichfork(flags
);
4092 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
|XFS_BMAPI_ENTIRE
|
4093 XFS_BMAPI_IGSTATE
|XFS_BMAPI_COWFORK
)));
4094 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
));
4096 if (unlikely(XFS_TEST_ERROR(
4097 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4098 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4099 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4100 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW
, mp
);
4101 return -EFSCORRUPTED
;
4104 if (XFS_FORCED_SHUTDOWN(mp
))
4107 XFS_STATS_INC(mp
, xs_blk_mapr
);
4109 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4111 /* No CoW fork? Return a hole. */
4112 if (whichfork
== XFS_COW_FORK
&& !ifp
) {
4113 mval
->br_startoff
= bno
;
4114 mval
->br_startblock
= HOLESTARTBLOCK
;
4115 mval
->br_blockcount
= len
;
4116 mval
->br_state
= XFS_EXT_NORM
;
4121 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4122 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
4127 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &idx
, &got
))
4132 while (bno
< end
&& n
< *nmap
) {
4133 /* Reading past eof, act as though there's a hole up to end. */
4135 got
.br_startoff
= end
;
4136 if (got
.br_startoff
> bno
) {
4137 /* Reading in a hole. */
4138 mval
->br_startoff
= bno
;
4139 mval
->br_startblock
= HOLESTARTBLOCK
;
4140 mval
->br_blockcount
=
4141 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
4142 mval
->br_state
= XFS_EXT_NORM
;
4143 bno
+= mval
->br_blockcount
;
4144 len
-= mval
->br_blockcount
;
4150 /* set up the extent map to return. */
4151 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4152 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4154 /* If we're done, stop now. */
4155 if (bno
>= end
|| n
>= *nmap
)
4158 /* Else go on to the next record. */
4159 if (!xfs_iext_get_extent(ifp
, ++idx
, &got
))
4167 xfs_bmapi_reserve_delalloc(
4168 struct xfs_inode
*ip
,
4172 xfs_filblks_t prealloc
,
4173 struct xfs_bmbt_irec
*got
,
4174 xfs_extnum_t
*lastx
,
4177 struct xfs_mount
*mp
= ip
->i_mount
;
4178 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4180 xfs_extlen_t indlen
;
4181 char rt
= XFS_IS_REALTIME_INODE(ip
);
4184 xfs_fileoff_t aoff
= off
;
4187 * Cap the alloc length. Keep track of prealloc so we know whether to
4188 * tag the inode before we return.
4190 alen
= XFS_FILBLKS_MIN(len
+ prealloc
, MAXEXTLEN
);
4192 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
4193 if (prealloc
&& alen
>= len
)
4194 prealloc
= alen
- len
;
4196 /* Figure out the extent size, adjust alen */
4197 if (whichfork
== XFS_COW_FORK
)
4198 extsz
= xfs_get_cowextsz_hint(ip
);
4200 extsz
= xfs_get_extsz_hint(ip
);
4202 struct xfs_bmbt_irec prev
;
4204 if (!xfs_iext_get_extent(ifp
, *lastx
- 1, &prev
))
4205 prev
.br_startoff
= NULLFILEOFF
;
4207 error
= xfs_bmap_extsize_align(mp
, got
, &prev
, extsz
, rt
, eof
,
4208 1, 0, &aoff
, &alen
);
4213 extsz
= alen
/ mp
->m_sb
.sb_rextsize
;
4216 * Make a transaction-less quota reservation for delayed allocation
4217 * blocks. This number gets adjusted later. We return if we haven't
4218 * allocated blocks already inside this loop.
4220 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4221 rt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4226 * Split changing sb for alen and indlen since they could be coming
4227 * from different places.
4229 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
4233 error
= xfs_mod_frextents(mp
, -((int64_t)extsz
));
4235 error
= xfs_mod_fdblocks(mp
, -((int64_t)alen
), false);
4239 goto out_unreserve_quota
;
4241 error
= xfs_mod_fdblocks(mp
, -((int64_t)indlen
), false);
4243 goto out_unreserve_blocks
;
4246 ip
->i_delayed_blks
+= alen
;
4248 got
->br_startoff
= aoff
;
4249 got
->br_startblock
= nullstartblock(indlen
);
4250 got
->br_blockcount
= alen
;
4251 got
->br_state
= XFS_EXT_NORM
;
4252 xfs_bmap_add_extent_hole_delay(ip
, whichfork
, lastx
, got
);
4255 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4256 * might have merged it into one of the neighbouring ones.
4258 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *lastx
), got
);
4261 * Tag the inode if blocks were preallocated. Note that COW fork
4262 * preallocation can occur at the start or end of the extent, even when
4263 * prealloc == 0, so we must also check the aligned offset and length.
4265 if (whichfork
== XFS_DATA_FORK
&& prealloc
)
4266 xfs_inode_set_eofblocks_tag(ip
);
4267 if (whichfork
== XFS_COW_FORK
&& (prealloc
|| aoff
< off
|| alen
> len
))
4268 xfs_inode_set_cowblocks_tag(ip
);
4270 ASSERT(got
->br_startoff
<= aoff
);
4271 ASSERT(got
->br_startoff
+ got
->br_blockcount
>= aoff
+ alen
);
4272 ASSERT(isnullstartblock(got
->br_startblock
));
4273 ASSERT(got
->br_state
== XFS_EXT_NORM
);
4276 out_unreserve_blocks
:
4278 xfs_mod_frextents(mp
, extsz
);
4280 xfs_mod_fdblocks(mp
, alen
, false);
4281 out_unreserve_quota
:
4282 if (XFS_IS_QUOTA_ON(mp
))
4283 xfs_trans_unreserve_quota_nblks(NULL
, ip
, (long)alen
, 0, rt
?
4284 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4290 struct xfs_bmalloca
*bma
)
4292 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4293 int whichfork
= xfs_bmapi_whichfork(bma
->flags
);
4294 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4295 int tmp_logflags
= 0;
4298 ASSERT(bma
->length
> 0);
4301 * For the wasdelay case, we could also just allocate the stuff asked
4302 * for in this bmap call but that wouldn't be as good.
4305 bma
->length
= (xfs_extlen_t
)bma
->got
.br_blockcount
;
4306 bma
->offset
= bma
->got
.br_startoff
;
4308 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
4312 bma
->length
= XFS_FILBLKS_MIN(bma
->length
, MAXEXTLEN
);
4314 bma
->length
= XFS_FILBLKS_MIN(bma
->length
,
4315 bma
->got
.br_startoff
- bma
->offset
);
4319 * Set the data type being allocated. For the data fork, the first data
4320 * in the file is treated differently to all other allocations. For the
4321 * attribute fork, we only need to ensure the allocated range is not on
4324 if (!(bma
->flags
& XFS_BMAPI_METADATA
)) {
4325 bma
->datatype
= XFS_ALLOC_NOBUSY
;
4326 if (whichfork
== XFS_DATA_FORK
) {
4327 if (bma
->offset
== 0)
4328 bma
->datatype
|= XFS_ALLOC_INITIAL_USER_DATA
;
4330 bma
->datatype
|= XFS_ALLOC_USERDATA
;
4332 if (bma
->flags
& XFS_BMAPI_ZERO
)
4333 bma
->datatype
|= XFS_ALLOC_USERDATA_ZERO
;
4336 bma
->minlen
= (bma
->flags
& XFS_BMAPI_CONTIG
) ? bma
->length
: 1;
4339 * Only want to do the alignment at the eof if it is userdata and
4340 * allocation length is larger than a stripe unit.
4342 if (mp
->m_dalign
&& bma
->length
>= mp
->m_dalign
&&
4343 !(bma
->flags
& XFS_BMAPI_METADATA
) && whichfork
== XFS_DATA_FORK
) {
4344 error
= xfs_bmap_isaeof(bma
, whichfork
);
4349 error
= xfs_bmap_alloc(bma
);
4354 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4355 if (bma
->blkno
== NULLFSBLOCK
)
4357 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4358 bma
->cur
= xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4359 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4360 bma
->cur
->bc_private
.b
.dfops
= bma
->dfops
;
4363 * Bump the number of extents we've allocated
4369 bma
->cur
->bc_private
.b
.flags
=
4370 bma
->wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
4372 bma
->got
.br_startoff
= bma
->offset
;
4373 bma
->got
.br_startblock
= bma
->blkno
;
4374 bma
->got
.br_blockcount
= bma
->length
;
4375 bma
->got
.br_state
= XFS_EXT_NORM
;
4378 * In the data fork, a wasdelay extent has been initialized, so
4379 * shouldn't be flagged as unwritten.
4381 * For the cow fork, however, we convert delalloc reservations
4382 * (extents allocated for speculative preallocation) to
4383 * allocated unwritten extents, and only convert the unwritten
4384 * extents to real extents when we're about to write the data.
4386 if ((!bma
->wasdel
|| (bma
->flags
& XFS_BMAPI_COWFORK
)) &&
4387 (bma
->flags
& XFS_BMAPI_PREALLOC
) &&
4388 xfs_sb_version_hasextflgbit(&mp
->m_sb
))
4389 bma
->got
.br_state
= XFS_EXT_UNWRITTEN
;
4392 error
= xfs_bmap_add_extent_delay_real(bma
, whichfork
);
4394 error
= xfs_bmap_add_extent_hole_real(bma
, whichfork
);
4396 bma
->logflags
|= tmp_logflags
;
4401 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4402 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4403 * the neighbouring ones.
4405 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4407 ASSERT(bma
->got
.br_startoff
<= bma
->offset
);
4408 ASSERT(bma
->got
.br_startoff
+ bma
->got
.br_blockcount
>=
4409 bma
->offset
+ bma
->length
);
4410 ASSERT(bma
->got
.br_state
== XFS_EXT_NORM
||
4411 bma
->got
.br_state
== XFS_EXT_UNWRITTEN
);
4416 xfs_bmapi_convert_unwritten(
4417 struct xfs_bmalloca
*bma
,
4418 struct xfs_bmbt_irec
*mval
,
4422 int whichfork
= xfs_bmapi_whichfork(flags
);
4423 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4424 int tmp_logflags
= 0;
4427 /* check if we need to do unwritten->real conversion */
4428 if (mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4429 (flags
& XFS_BMAPI_PREALLOC
))
4432 /* check if we need to do real->unwritten conversion */
4433 if (mval
->br_state
== XFS_EXT_NORM
&&
4434 (flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
)) !=
4435 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
))
4439 * Modify (by adding) the state flag, if writing.
4441 ASSERT(mval
->br_blockcount
<= len
);
4442 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4443 bma
->cur
= xfs_bmbt_init_cursor(bma
->ip
->i_mount
, bma
->tp
,
4444 bma
->ip
, whichfork
);
4445 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4446 bma
->cur
->bc_private
.b
.dfops
= bma
->dfops
;
4448 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4449 ? XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
4452 * Before insertion into the bmbt, zero the range being converted
4455 if (flags
& XFS_BMAPI_ZERO
) {
4456 error
= xfs_zero_extent(bma
->ip
, mval
->br_startblock
,
4457 mval
->br_blockcount
);
4462 error
= xfs_bmap_add_extent_unwritten_real(bma
->tp
, bma
->ip
, whichfork
,
4463 &bma
->idx
, &bma
->cur
, mval
, bma
->firstblock
, bma
->dfops
,
4466 * Log the inode core unconditionally in the unwritten extent conversion
4467 * path because the conversion might not have done so (e.g., if the
4468 * extent count hasn't changed). We need to make sure the inode is dirty
4469 * in the transaction for the sake of fsync(), even if nothing has
4470 * changed, because fsync() will not force the log for this transaction
4471 * unless it sees the inode pinned.
4473 * Note: If we're only converting cow fork extents, there aren't
4474 * any on-disk updates to make, so we don't need to log anything.
4476 if (whichfork
!= XFS_COW_FORK
)
4477 bma
->logflags
|= tmp_logflags
| XFS_ILOG_CORE
;
4482 * Update our extent pointer, given that
4483 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4484 * of the neighbouring ones.
4486 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4489 * We may have combined previously unwritten space with written space,
4490 * so generate another request.
4492 if (mval
->br_blockcount
< len
)
4498 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4499 * extent state if necessary. Details behaviour is controlled by the flags
4500 * parameter. Only allocates blocks from a single allocation group, to avoid
4503 * The returned value in "firstblock" from the first call in a transaction
4504 * must be remembered and presented to subsequent calls in "firstblock".
4505 * An upper bound for the number of blocks to be allocated is supplied to
4506 * the first call in "total"; if no allocation group has that many free
4507 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4511 struct xfs_trans
*tp
, /* transaction pointer */
4512 struct xfs_inode
*ip
, /* incore inode */
4513 xfs_fileoff_t bno
, /* starting file offs. mapped */
4514 xfs_filblks_t len
, /* length to map in file */
4515 int flags
, /* XFS_BMAPI_... */
4516 xfs_fsblock_t
*firstblock
, /* first allocated block
4517 controls a.g. for allocs */
4518 xfs_extlen_t total
, /* total blocks needed */
4519 struct xfs_bmbt_irec
*mval
, /* output: map values */
4520 int *nmap
, /* i/o: mval size/count */
4521 struct xfs_defer_ops
*dfops
) /* i/o: list extents to free */
4523 struct xfs_mount
*mp
= ip
->i_mount
;
4524 struct xfs_ifork
*ifp
;
4525 struct xfs_bmalloca bma
= { NULL
}; /* args for xfs_bmap_alloc */
4526 xfs_fileoff_t end
; /* end of mapped file region */
4527 bool eof
= false; /* after the end of extents */
4528 int error
; /* error return */
4529 int n
; /* current extent index */
4530 xfs_fileoff_t obno
; /* old block number (offset) */
4531 int whichfork
; /* data or attr fork */
4534 xfs_fileoff_t orig_bno
; /* original block number value */
4535 int orig_flags
; /* original flags arg value */
4536 xfs_filblks_t orig_len
; /* original value of len arg */
4537 struct xfs_bmbt_irec
*orig_mval
; /* original value of mval */
4538 int orig_nmap
; /* original value of *nmap */
4546 whichfork
= xfs_bmapi_whichfork(flags
);
4549 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4550 ASSERT(!(flags
& XFS_BMAPI_IGSTATE
));
4551 ASSERT(tp
!= NULL
||
4552 (flags
& (XFS_BMAPI_CONVERT
| XFS_BMAPI_COWFORK
)) ==
4553 (XFS_BMAPI_CONVERT
| XFS_BMAPI_COWFORK
));
4555 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
);
4556 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4557 ASSERT(!(flags
& XFS_BMAPI_REMAP
) || whichfork
== XFS_DATA_FORK
);
4558 ASSERT(!(flags
& XFS_BMAPI_PREALLOC
) || !(flags
& XFS_BMAPI_REMAP
));
4559 ASSERT(!(flags
& XFS_BMAPI_CONVERT
) || !(flags
& XFS_BMAPI_REMAP
));
4561 /* zeroing is for currently only for data extents, not metadata */
4562 ASSERT((flags
& (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
)) !=
4563 (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
));
4565 * we can allocate unwritten extents or pre-zero allocated blocks,
4566 * but it makes no sense to do both at once. This would result in
4567 * zeroing the unwritten extent twice, but it still being an
4568 * unwritten extent....
4570 ASSERT((flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
)) !=
4571 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
));
4573 if (unlikely(XFS_TEST_ERROR(
4574 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4575 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4576 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4577 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW
, mp
);
4578 return -EFSCORRUPTED
;
4581 if (XFS_FORCED_SHUTDOWN(mp
))
4584 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4586 XFS_STATS_INC(mp
, xs_blk_mapw
);
4588 if (*firstblock
== NULLFSBLOCK
) {
4589 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
)
4590 bma
.minleft
= be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
4597 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4598 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4607 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &bma
.idx
, &bma
.got
))
4609 if (!xfs_iext_get_extent(ifp
, bma
.idx
- 1, &bma
.prev
))
4610 bma
.prev
.br_startoff
= NULLFILEOFF
;
4616 bma
.firstblock
= firstblock
;
4618 while (bno
< end
&& n
< *nmap
) {
4619 bool need_alloc
= false, wasdelay
= false;
4621 /* in hole or beyoned EOF? */
4622 if (eof
|| bma
.got
.br_startoff
> bno
) {
4623 if (flags
& XFS_BMAPI_DELALLOC
) {
4625 * For the COW fork we can reasonably get a
4626 * request for converting an extent that races
4627 * with other threads already having converted
4628 * part of it, as there converting COW to
4629 * regular blocks is not protected using the
4632 ASSERT(flags
& XFS_BMAPI_COWFORK
);
4633 if (!(flags
& XFS_BMAPI_COWFORK
)) {
4638 if (eof
|| bno
>= end
)
4645 * Make sure we only reflink into a hole.
4647 ASSERT(!(flags
& XFS_BMAPI_REMAP
));
4648 if (isnullstartblock(bma
.got
.br_startblock
))
4653 * First, deal with the hole before the allocated space
4654 * that we found, if any.
4656 if (need_alloc
|| wasdelay
) {
4658 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
4659 bma
.wasdel
= wasdelay
;
4664 * There's a 32/64 bit type mismatch between the
4665 * allocation length request (which can be 64 bits in
4666 * length) and the bma length request, which is
4667 * xfs_extlen_t and therefore 32 bits. Hence we have to
4668 * check for 32-bit overflows and handle them here.
4670 if (len
> (xfs_filblks_t
)MAXEXTLEN
)
4671 bma
.length
= MAXEXTLEN
;
4676 ASSERT(bma
.length
> 0);
4677 error
= xfs_bmapi_allocate(&bma
);
4680 if (bma
.blkno
== NULLFSBLOCK
)
4684 * If this is a CoW allocation, record the data in
4685 * the refcount btree for orphan recovery.
4687 if (whichfork
== XFS_COW_FORK
) {
4688 error
= xfs_refcount_alloc_cow_extent(mp
, dfops
,
4689 bma
.blkno
, bma
.length
);
4695 /* Deal with the allocated space we found. */
4696 xfs_bmapi_trim_map(mval
, &bma
.got
, &bno
, len
, obno
,
4699 /* Execute unwritten extent conversion if necessary */
4700 error
= xfs_bmapi_convert_unwritten(&bma
, mval
, len
, flags
);
4701 if (error
== -EAGAIN
)
4706 /* update the extent map to return */
4707 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4710 * If we're done, stop now. Stop when we've allocated
4711 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4712 * the transaction may get too big.
4714 if (bno
>= end
|| n
>= *nmap
|| bma
.nallocs
>= *nmap
)
4717 /* Else go on to the next record. */
4719 if (!xfs_iext_get_extent(ifp
, ++bma
.idx
, &bma
.got
))
4725 * Transform from btree to extents, give it cur.
4727 if (xfs_bmap_wants_extents(ip
, whichfork
)) {
4728 int tmp_logflags
= 0;
4731 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
,
4732 &tmp_logflags
, whichfork
);
4733 bma
.logflags
|= tmp_logflags
;
4738 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
||
4739 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
4740 XFS_IFORK_MAXEXT(ip
, whichfork
));
4744 * Log everything. Do this after conversion, there's no point in
4745 * logging the extent records if we've converted to btree format.
4747 if ((bma
.logflags
& xfs_ilog_fext(whichfork
)) &&
4748 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
4749 bma
.logflags
&= ~xfs_ilog_fext(whichfork
);
4750 else if ((bma
.logflags
& xfs_ilog_fbroot(whichfork
)) &&
4751 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
4752 bma
.logflags
&= ~xfs_ilog_fbroot(whichfork
);
4754 * Log whatever the flags say, even if error. Otherwise we might miss
4755 * detecting a case where the data is changed, there's an error,
4756 * and it's not logged so we don't shutdown when we should.
4759 xfs_trans_log_inode(tp
, ip
, bma
.logflags
);
4763 ASSERT(*firstblock
== NULLFSBLOCK
||
4764 XFS_FSB_TO_AGNO(mp
, *firstblock
) ==
4766 bma
.cur
->bc_private
.b
.firstblock
) ||
4768 XFS_FSB_TO_AGNO(mp
, *firstblock
) <
4770 bma
.cur
->bc_private
.b
.firstblock
)));
4771 *firstblock
= bma
.cur
->bc_private
.b
.firstblock
;
4773 xfs_btree_del_cursor(bma
.cur
,
4774 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
4777 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
4783 * When a delalloc extent is split (e.g., due to a hole punch), the original
4784 * indlen reservation must be shared across the two new extents that are left
4787 * Given the original reservation and the worst case indlen for the two new
4788 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4789 * reservation fairly across the two new extents. If necessary, steal available
4790 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4791 * ores == 1). The number of stolen blocks is returned. The availability and
4792 * subsequent accounting of stolen blocks is the responsibility of the caller.
4794 static xfs_filblks_t
4795 xfs_bmap_split_indlen(
4796 xfs_filblks_t ores
, /* original res. */
4797 xfs_filblks_t
*indlen1
, /* ext1 worst indlen */
4798 xfs_filblks_t
*indlen2
, /* ext2 worst indlen */
4799 xfs_filblks_t avail
) /* stealable blocks */
4801 xfs_filblks_t len1
= *indlen1
;
4802 xfs_filblks_t len2
= *indlen2
;
4803 xfs_filblks_t nres
= len1
+ len2
; /* new total res. */
4804 xfs_filblks_t stolen
= 0;
4807 * Steal as many blocks as we can to try and satisfy the worst case
4808 * indlen for both new extents.
4810 while (nres
> ores
&& avail
) {
4817 * The only blocks available are those reserved for the original
4818 * extent and what we can steal from the extent being removed.
4819 * If this still isn't enough to satisfy the combined
4820 * requirements for the two new extents, skim blocks off of each
4821 * of the new reservations until they match what is available.
4823 while (nres
> ores
) {
4843 xfs_bmap_del_extent_delay(
4844 struct xfs_inode
*ip
,
4847 struct xfs_bmbt_irec
*got
,
4848 struct xfs_bmbt_irec
*del
)
4850 struct xfs_mount
*mp
= ip
->i_mount
;
4851 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4852 struct xfs_bmbt_irec
new;
4853 int64_t da_old
, da_new
, da_diff
= 0;
4854 xfs_fileoff_t del_endoff
, got_endoff
;
4855 xfs_filblks_t got_indlen
, new_indlen
, stolen
;
4856 int error
= 0, state
= 0;
4859 XFS_STATS_INC(mp
, xs_del_exlist
);
4861 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
4862 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4863 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4864 da_old
= startblockval(got
->br_startblock
);
4868 ASSERT(*idx
<= xfs_iext_count(ifp
));
4869 ASSERT(del
->br_blockcount
> 0);
4870 ASSERT(got
->br_startoff
<= del
->br_startoff
);
4871 ASSERT(got_endoff
>= del_endoff
);
4874 int64_t rtexts
= XFS_FSB_TO_B(mp
, del
->br_blockcount
);
4876 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
4877 xfs_mod_frextents(mp
, rtexts
);
4881 * Update the inode delalloc counter now and wait to update the
4882 * sb counters as we might have to borrow some blocks for the
4883 * indirect block accounting.
4885 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
,
4886 -((long)del
->br_blockcount
), 0,
4887 isrt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4890 ip
->i_delayed_blks
-= del
->br_blockcount
;
4892 if (whichfork
== XFS_COW_FORK
)
4893 state
|= BMAP_COWFORK
;
4895 if (got
->br_startoff
== del
->br_startoff
)
4896 state
|= BMAP_LEFT_CONTIG
;
4897 if (got_endoff
== del_endoff
)
4898 state
|= BMAP_RIGHT_CONTIG
;
4900 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
4901 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
4903 * Matches the whole extent. Delete the entry.
4905 xfs_iext_remove(ip
, *idx
, 1, state
);
4908 case BMAP_LEFT_CONTIG
:
4910 * Deleting the first part of the extent.
4912 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
4913 got
->br_startoff
= del_endoff
;
4914 got
->br_blockcount
-= del
->br_blockcount
;
4915 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4916 got
->br_blockcount
), da_old
);
4917 got
->br_startblock
= nullstartblock((int)da_new
);
4918 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
4919 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
4921 case BMAP_RIGHT_CONTIG
:
4923 * Deleting the last part of the extent.
4925 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
4926 got
->br_blockcount
= got
->br_blockcount
- del
->br_blockcount
;
4927 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4928 got
->br_blockcount
), da_old
);
4929 got
->br_startblock
= nullstartblock((int)da_new
);
4930 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
4931 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
4935 * Deleting the middle of the extent.
4937 * Distribute the original indlen reservation across the two new
4938 * extents. Steal blocks from the deleted extent if necessary.
4939 * Stealing blocks simply fudges the fdblocks accounting below.
4940 * Warn if either of the new indlen reservations is zero as this
4941 * can lead to delalloc problems.
4943 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
4945 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
4946 got_indlen
= xfs_bmap_worst_indlen(ip
, got
->br_blockcount
);
4948 new.br_blockcount
= got_endoff
- del_endoff
;
4949 new_indlen
= xfs_bmap_worst_indlen(ip
, new.br_blockcount
);
4951 WARN_ON_ONCE(!got_indlen
|| !new_indlen
);
4952 stolen
= xfs_bmap_split_indlen(da_old
, &got_indlen
, &new_indlen
,
4953 del
->br_blockcount
);
4955 got
->br_startblock
= nullstartblock((int)got_indlen
);
4956 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
4957 trace_xfs_bmap_post_update(ip
, *idx
, 0, _THIS_IP_
);
4959 new.br_startoff
= del_endoff
;
4960 new.br_state
= got
->br_state
;
4961 new.br_startblock
= nullstartblock((int)new_indlen
);
4964 xfs_iext_insert(ip
, *idx
, 1, &new, state
);
4966 da_new
= got_indlen
+ new_indlen
- stolen
;
4967 del
->br_blockcount
-= stolen
;
4971 ASSERT(da_old
>= da_new
);
4972 da_diff
= da_old
- da_new
;
4974 da_diff
+= del
->br_blockcount
;
4976 xfs_mod_fdblocks(mp
, da_diff
, false);
4981 xfs_bmap_del_extent_cow(
4982 struct xfs_inode
*ip
,
4984 struct xfs_bmbt_irec
*got
,
4985 struct xfs_bmbt_irec
*del
)
4987 struct xfs_mount
*mp
= ip
->i_mount
;
4988 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
4989 struct xfs_bmbt_irec
new;
4990 xfs_fileoff_t del_endoff
, got_endoff
;
4991 int state
= BMAP_COWFORK
;
4993 XFS_STATS_INC(mp
, xs_del_exlist
);
4995 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4996 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4999 ASSERT(*idx
<= xfs_iext_count(ifp
));
5000 ASSERT(del
->br_blockcount
> 0);
5001 ASSERT(got
->br_startoff
<= del
->br_startoff
);
5002 ASSERT(got_endoff
>= del_endoff
);
5003 ASSERT(!isnullstartblock(got
->br_startblock
));
5005 if (got
->br_startoff
== del
->br_startoff
)
5006 state
|= BMAP_LEFT_CONTIG
;
5007 if (got_endoff
== del_endoff
)
5008 state
|= BMAP_RIGHT_CONTIG
;
5010 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
5011 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
5013 * Matches the whole extent. Delete the entry.
5015 xfs_iext_remove(ip
, *idx
, 1, state
);
5018 case BMAP_LEFT_CONTIG
:
5020 * Deleting the first part of the extent.
5022 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5023 got
->br_startoff
= del_endoff
;
5024 got
->br_blockcount
-= del
->br_blockcount
;
5025 got
->br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
5026 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5027 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5029 case BMAP_RIGHT_CONTIG
:
5031 * Deleting the last part of the extent.
5033 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5034 got
->br_blockcount
-= del
->br_blockcount
;
5035 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5036 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5040 * Deleting the middle of the extent.
5042 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5043 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
5044 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5045 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5047 new.br_startoff
= del_endoff
;
5048 new.br_blockcount
= got_endoff
- del_endoff
;
5049 new.br_state
= got
->br_state
;
5050 new.br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
5053 xfs_iext_insert(ip
, *idx
, 1, &new, state
);
5059 * Called by xfs_bmapi to update file extent records and the btree
5060 * after removing space (or undoing a delayed allocation).
5062 STATIC
int /* error */
5063 xfs_bmap_del_extent(
5064 xfs_inode_t
*ip
, /* incore inode pointer */
5065 xfs_trans_t
*tp
, /* current transaction pointer */
5066 xfs_extnum_t
*idx
, /* extent number to update/delete */
5067 struct xfs_defer_ops
*dfops
, /* list of extents to be freed */
5068 xfs_btree_cur_t
*cur
, /* if null, not a btree */
5069 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
5070 int *logflagsp
, /* inode logging flags */
5071 int whichfork
, /* data or attr fork */
5072 int bflags
) /* bmapi flags */
5074 xfs_filblks_t da_new
; /* new delay-alloc indirect blocks */
5075 xfs_filblks_t da_old
; /* old delay-alloc indirect blocks */
5076 xfs_fsblock_t del_endblock
=0; /* first block past del */
5077 xfs_fileoff_t del_endoff
; /* first offset past del */
5078 int delay
; /* current block is delayed allocated */
5079 int do_fx
; /* free extent at end of routine */
5080 xfs_bmbt_rec_host_t
*ep
; /* current extent entry pointer */
5081 int error
; /* error return value */
5082 int flags
; /* inode logging flags */
5083 xfs_bmbt_irec_t got
; /* current extent entry */
5084 xfs_fileoff_t got_endoff
; /* first offset past got */
5085 int i
; /* temp state */
5086 xfs_ifork_t
*ifp
; /* inode fork pointer */
5087 xfs_mount_t
*mp
; /* mount structure */
5088 xfs_filblks_t nblks
; /* quota/sb block count */
5089 xfs_bmbt_irec_t
new; /* new record to be inserted */
5091 uint qfield
; /* quota field to update */
5092 xfs_filblks_t temp
; /* for indirect length calculations */
5093 xfs_filblks_t temp2
; /* for indirect length calculations */
5097 XFS_STATS_INC(mp
, xs_del_exlist
);
5099 if (whichfork
== XFS_ATTR_FORK
)
5100 state
|= BMAP_ATTRFORK
;
5101 else if (whichfork
== XFS_COW_FORK
)
5102 state
|= BMAP_COWFORK
;
5104 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5105 ASSERT((*idx
>= 0) && (*idx
< xfs_iext_count(ifp
)));
5106 ASSERT(del
->br_blockcount
> 0);
5107 ep
= xfs_iext_get_ext(ifp
, *idx
);
5108 xfs_bmbt_get_all(ep
, &got
);
5109 ASSERT(got
.br_startoff
<= del
->br_startoff
);
5110 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
5111 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
5112 ASSERT(got_endoff
>= del_endoff
);
5113 delay
= isnullstartblock(got
.br_startblock
);
5114 ASSERT(isnullstartblock(del
->br_startblock
) == delay
);
5119 * If deleting a real allocation, must free up the disk space.
5122 flags
= XFS_ILOG_CORE
;
5124 * Realtime allocation. Free it and record di_nblocks update.
5126 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
5130 ASSERT(do_mod(del
->br_blockcount
,
5131 mp
->m_sb
.sb_rextsize
) == 0);
5132 ASSERT(do_mod(del
->br_startblock
,
5133 mp
->m_sb
.sb_rextsize
) == 0);
5134 bno
= del
->br_startblock
;
5135 len
= del
->br_blockcount
;
5136 do_div(bno
, mp
->m_sb
.sb_rextsize
);
5137 do_div(len
, mp
->m_sb
.sb_rextsize
);
5138 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
5142 nblks
= len
* mp
->m_sb
.sb_rextsize
;
5143 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
5146 * Ordinary allocation.
5150 nblks
= del
->br_blockcount
;
5151 qfield
= XFS_TRANS_DQ_BCOUNT
;
5154 * Set up del_endblock and cur for later.
5156 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
5158 if ((error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
5159 got
.br_startblock
, got
.br_blockcount
,
5162 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5164 da_old
= da_new
= 0;
5166 da_old
= startblockval(got
.br_startblock
);
5173 * Set flag value to use in switch statement.
5174 * Left-contig is 2, right-contig is 1.
5176 switch (((got
.br_startoff
== del
->br_startoff
) << 1) |
5177 (got_endoff
== del_endoff
)) {
5180 * Matches the whole extent. Delete the entry.
5182 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5183 xfs_iext_remove(ip
, *idx
, 1,
5184 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0);
5189 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5190 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
5191 flags
|= XFS_ILOG_CORE
;
5193 flags
|= xfs_ilog_fext(whichfork
);
5196 if ((error
= xfs_btree_delete(cur
, &i
)))
5198 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5203 * Deleting the first part of the extent.
5205 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5206 xfs_bmbt_set_startoff(ep
, del_endoff
);
5207 temp
= got
.br_blockcount
- del
->br_blockcount
;
5208 xfs_bmbt_set_blockcount(ep
, temp
);
5210 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
5212 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
5213 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5217 xfs_bmbt_set_startblock(ep
, del_endblock
);
5218 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5220 flags
|= xfs_ilog_fext(whichfork
);
5223 if ((error
= xfs_bmbt_update(cur
, del_endoff
, del_endblock
,
5224 got
.br_blockcount
- del
->br_blockcount
,
5231 * Deleting the last part of the extent.
5233 temp
= got
.br_blockcount
- del
->br_blockcount
;
5234 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5235 xfs_bmbt_set_blockcount(ep
, temp
);
5237 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
5239 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
5240 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5244 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5246 flags
|= xfs_ilog_fext(whichfork
);
5249 if ((error
= xfs_bmbt_update(cur
, got
.br_startoff
,
5251 got
.br_blockcount
- del
->br_blockcount
,
5258 * Deleting the middle of the extent.
5260 temp
= del
->br_startoff
- got
.br_startoff
;
5261 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5262 xfs_bmbt_set_blockcount(ep
, temp
);
5263 new.br_startoff
= del_endoff
;
5264 temp2
= got_endoff
- del_endoff
;
5265 new.br_blockcount
= temp2
;
5266 new.br_state
= got
.br_state
;
5268 new.br_startblock
= del_endblock
;
5269 flags
|= XFS_ILOG_CORE
;
5271 if ((error
= xfs_bmbt_update(cur
,
5273 got
.br_startblock
, temp
,
5276 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
5278 cur
->bc_rec
.b
= new;
5279 error
= xfs_btree_insert(cur
, &i
);
5280 if (error
&& error
!= -ENOSPC
)
5283 * If get no-space back from btree insert,
5284 * it tried a split, and we have a zero
5285 * block reservation.
5286 * Fix up our state and return the error.
5288 if (error
== -ENOSPC
) {
5290 * Reset the cursor, don't trust
5291 * it after any insert operation.
5293 if ((error
= xfs_bmbt_lookup_eq(cur
,
5298 XFS_WANT_CORRUPTED_GOTO(mp
,
5301 * Update the btree record back
5302 * to the original value.
5304 if ((error
= xfs_bmbt_update(cur
,
5311 * Reset the extent record back
5312 * to the original value.
5314 xfs_bmbt_set_blockcount(ep
,
5320 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5322 flags
|= xfs_ilog_fext(whichfork
);
5323 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5324 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
5326 xfs_filblks_t stolen
;
5327 ASSERT(whichfork
== XFS_DATA_FORK
);
5330 * Distribute the original indlen reservation across the
5331 * two new extents. Steal blocks from the deleted extent
5332 * if necessary. Stealing blocks simply fudges the
5333 * fdblocks accounting in xfs_bunmapi().
5335 temp
= xfs_bmap_worst_indlen(ip
, got
.br_blockcount
);
5336 temp2
= xfs_bmap_worst_indlen(ip
, new.br_blockcount
);
5337 stolen
= xfs_bmap_split_indlen(da_old
, &temp
, &temp2
,
5338 del
->br_blockcount
);
5339 da_new
= temp
+ temp2
- stolen
;
5340 del
->br_blockcount
-= stolen
;
5343 * Set the reservation for each extent. Warn if either
5344 * is zero as this can lead to delalloc problems.
5346 WARN_ON_ONCE(!temp
|| !temp2
);
5347 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
5348 new.br_startblock
= nullstartblock((int)temp2
);
5350 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5351 xfs_iext_insert(ip
, *idx
+ 1, 1, &new, state
);
5356 /* remove reverse mapping */
5358 error
= xfs_rmap_unmap_extent(mp
, dfops
, ip
, whichfork
, del
);
5364 * If we need to, add to list of extents to delete.
5366 if (do_fx
&& !(bflags
& XFS_BMAPI_REMAP
)) {
5367 if (xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
) {
5368 error
= xfs_refcount_decrease_extent(mp
, dfops
, del
);
5372 xfs_bmap_add_free(mp
, dfops
, del
->br_startblock
,
5373 del
->br_blockcount
, NULL
);
5377 * Adjust inode # blocks in the file.
5380 ip
->i_d
.di_nblocks
-= nblks
;
5382 * Adjust quota data.
5384 if (qfield
&& !(bflags
& XFS_BMAPI_REMAP
))
5385 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
5388 * Account for change in delayed indirect blocks.
5389 * Nothing to do for disk quota accounting here.
5391 ASSERT(da_old
>= da_new
);
5392 if (da_old
> da_new
)
5393 xfs_mod_fdblocks(mp
, (int64_t)(da_old
- da_new
), false);
5400 * Unmap (remove) blocks from a file.
5401 * If nexts is nonzero then the number of extents to remove is limited to
5402 * that value. If not all extents in the block range can be removed then
5407 xfs_trans_t
*tp
, /* transaction pointer */
5408 struct xfs_inode
*ip
, /* incore inode */
5409 xfs_fileoff_t bno
, /* starting offset to unmap */
5410 xfs_filblks_t
*rlen
, /* i/o: amount remaining */
5411 int flags
, /* misc flags */
5412 xfs_extnum_t nexts
, /* number of extents max */
5413 xfs_fsblock_t
*firstblock
, /* first allocated block
5414 controls a.g. for allocs */
5415 struct xfs_defer_ops
*dfops
) /* i/o: deferred updates */
5417 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
5418 xfs_bmbt_irec_t del
; /* extent being deleted */
5419 int error
; /* error return value */
5420 xfs_extnum_t extno
; /* extent number in list */
5421 xfs_bmbt_irec_t got
; /* current extent record */
5422 xfs_ifork_t
*ifp
; /* inode fork pointer */
5423 int isrt
; /* freeing in rt area */
5424 xfs_extnum_t lastx
; /* last extent index used */
5425 int logflags
; /* transaction logging flags */
5426 xfs_extlen_t mod
; /* rt extent offset */
5427 xfs_mount_t
*mp
; /* mount structure */
5428 xfs_fileoff_t start
; /* first file offset deleted */
5429 int tmp_logflags
; /* partial logging flags */
5430 int wasdel
; /* was a delayed alloc extent */
5431 int whichfork
; /* data or attribute fork */
5433 xfs_filblks_t len
= *rlen
; /* length to unmap in file */
5435 trace_xfs_bunmap(ip
, bno
, len
, flags
, _RET_IP_
);
5437 whichfork
= xfs_bmapi_whichfork(flags
);
5438 ASSERT(whichfork
!= XFS_COW_FORK
);
5439 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5441 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5442 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
5443 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW
,
5445 return -EFSCORRUPTED
;
5448 if (XFS_FORCED_SHUTDOWN(mp
))
5451 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5455 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5456 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5458 if (xfs_iext_count(ifp
) == 0) {
5462 XFS_STATS_INC(mp
, xs_blk_unmap
);
5463 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5465 bno
= start
+ len
- 1;
5468 * Check to see if the given block number is past the end of the
5469 * file, back up to the last block if so...
5471 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &lastx
, &got
)) {
5473 xfs_iext_get_extent(ifp
, --lastx
, &got
);
5474 bno
= got
.br_startoff
+ got
.br_blockcount
- 1;
5478 if (ifp
->if_flags
& XFS_IFBROOT
) {
5479 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
5480 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5481 cur
->bc_private
.b
.firstblock
= *firstblock
;
5482 cur
->bc_private
.b
.dfops
= dfops
;
5483 cur
->bc_private
.b
.flags
= 0;
5489 * Synchronize by locking the bitmap inode.
5491 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTBITMAP
);
5492 xfs_trans_ijoin(tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5493 xfs_ilock(mp
->m_rsumip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTSUM
);
5494 xfs_trans_ijoin(tp
, mp
->m_rsumip
, XFS_ILOCK_EXCL
);
5498 while (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
&& lastx
>= 0 &&
5499 (nexts
== 0 || extno
< nexts
)) {
5501 * Is the found extent after a hole in which bno lives?
5502 * Just back up to the previous extent, if so.
5504 if (got
.br_startoff
> bno
) {
5507 xfs_iext_get_extent(ifp
, lastx
, &got
);
5510 * Is the last block of this extent before the range
5511 * we're supposed to delete? If so, we're done.
5513 bno
= XFS_FILEOFF_MIN(bno
,
5514 got
.br_startoff
+ got
.br_blockcount
- 1);
5518 * Then deal with the (possibly delayed) allocated space
5522 wasdel
= isnullstartblock(del
.br_startblock
);
5523 if (got
.br_startoff
< start
) {
5524 del
.br_startoff
= start
;
5525 del
.br_blockcount
-= start
- got
.br_startoff
;
5527 del
.br_startblock
+= start
- got
.br_startoff
;
5529 if (del
.br_startoff
+ del
.br_blockcount
> bno
+ 1)
5530 del
.br_blockcount
= bno
+ 1 - del
.br_startoff
;
5531 sum
= del
.br_startblock
+ del
.br_blockcount
;
5533 (mod
= do_mod(sum
, mp
->m_sb
.sb_rextsize
))) {
5535 * Realtime extent not lined up at the end.
5536 * The extent could have been split into written
5537 * and unwritten pieces, or we could just be
5538 * unmapping part of it. But we can't really
5539 * get rid of part of a realtime extent.
5541 if (del
.br_state
== XFS_EXT_UNWRITTEN
||
5542 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5544 * This piece is unwritten, or we're not
5545 * using unwritten extents. Skip over it.
5548 bno
-= mod
> del
.br_blockcount
?
5549 del
.br_blockcount
: mod
;
5550 if (bno
< got
.br_startoff
) {
5552 xfs_bmbt_get_all(xfs_iext_get_ext(
5558 * It's written, turn it unwritten.
5559 * This is better than zeroing it.
5561 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5562 ASSERT(tp
->t_blk_res
> 0);
5564 * If this spans a realtime extent boundary,
5565 * chop it back to the start of the one we end at.
5567 if (del
.br_blockcount
> mod
) {
5568 del
.br_startoff
+= del
.br_blockcount
- mod
;
5569 del
.br_startblock
+= del
.br_blockcount
- mod
;
5570 del
.br_blockcount
= mod
;
5572 del
.br_state
= XFS_EXT_UNWRITTEN
;
5573 error
= xfs_bmap_add_extent_unwritten_real(tp
, ip
,
5574 whichfork
, &lastx
, &cur
, &del
,
5575 firstblock
, dfops
, &logflags
);
5580 if (isrt
&& (mod
= do_mod(del
.br_startblock
, mp
->m_sb
.sb_rextsize
))) {
5582 * Realtime extent is lined up at the end but not
5583 * at the front. We'll get rid of full extents if
5586 mod
= mp
->m_sb
.sb_rextsize
- mod
;
5587 if (del
.br_blockcount
> mod
) {
5588 del
.br_blockcount
-= mod
;
5589 del
.br_startoff
+= mod
;
5590 del
.br_startblock
+= mod
;
5591 } else if ((del
.br_startoff
== start
&&
5592 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5593 tp
->t_blk_res
== 0)) ||
5594 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5596 * Can't make it unwritten. There isn't
5597 * a full extent here so just skip it.
5599 ASSERT(bno
>= del
.br_blockcount
);
5600 bno
-= del
.br_blockcount
;
5601 if (got
.br_startoff
> bno
&& --lastx
>= 0)
5602 xfs_iext_get_extent(ifp
, lastx
, &got
);
5604 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5605 struct xfs_bmbt_irec prev
;
5608 * This one is already unwritten.
5609 * It must have a written left neighbor.
5610 * Unwrite the killed part of that one and
5614 xfs_iext_get_extent(ifp
, lastx
- 1, &prev
);
5615 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5616 ASSERT(!isnullstartblock(prev
.br_startblock
));
5617 ASSERT(del
.br_startblock
==
5618 prev
.br_startblock
+ prev
.br_blockcount
);
5619 if (prev
.br_startoff
< start
) {
5620 mod
= start
- prev
.br_startoff
;
5621 prev
.br_blockcount
-= mod
;
5622 prev
.br_startblock
+= mod
;
5623 prev
.br_startoff
= start
;
5625 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5627 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5628 ip
, whichfork
, &lastx
, &cur
,
5629 &prev
, firstblock
, dfops
,
5635 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5636 del
.br_state
= XFS_EXT_UNWRITTEN
;
5637 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5638 ip
, whichfork
, &lastx
, &cur
,
5639 &del
, firstblock
, dfops
,
5648 * If it's the case where the directory code is running
5649 * with no block reservation, and the deleted block is in
5650 * the middle of its extent, and the resulting insert
5651 * of an extent would cause transformation to btree format,
5652 * then reject it. The calling code will then swap
5653 * blocks around instead.
5654 * We have to do this now, rather than waiting for the
5655 * conversion to btree format, since the transaction
5658 if (!wasdel
&& tp
->t_blk_res
== 0 &&
5659 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
5660 XFS_IFORK_NEXTENTS(ip
, whichfork
) >= /* Note the >= */
5661 XFS_IFORK_MAXEXT(ip
, whichfork
) &&
5662 del
.br_startoff
> got
.br_startoff
&&
5663 del
.br_startoff
+ del
.br_blockcount
<
5664 got
.br_startoff
+ got
.br_blockcount
) {
5670 * Unreserve quota and update realtime free space, if
5671 * appropriate. If delayed allocation, update the inode delalloc
5672 * counter now and wait to update the sb counters as
5673 * xfs_bmap_del_extent() might need to borrow some blocks.
5676 ASSERT(startblockval(del
.br_startblock
) > 0);
5678 xfs_filblks_t rtexts
;
5680 rtexts
= XFS_FSB_TO_B(mp
, del
.br_blockcount
);
5681 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
5682 xfs_mod_frextents(mp
, (int64_t)rtexts
);
5683 (void)xfs_trans_reserve_quota_nblks(NULL
,
5684 ip
, -((long)del
.br_blockcount
), 0,
5685 XFS_QMOPT_RES_RTBLKS
);
5687 (void)xfs_trans_reserve_quota_nblks(NULL
,
5688 ip
, -((long)del
.br_blockcount
), 0,
5689 XFS_QMOPT_RES_REGBLKS
);
5691 ip
->i_delayed_blks
-= del
.br_blockcount
;
5693 cur
->bc_private
.b
.flags
|=
5694 XFS_BTCUR_BPRV_WASDEL
;
5696 cur
->bc_private
.b
.flags
&= ~XFS_BTCUR_BPRV_WASDEL
;
5698 error
= xfs_bmap_del_extent(ip
, tp
, &lastx
, dfops
, cur
, &del
,
5699 &tmp_logflags
, whichfork
, flags
);
5700 logflags
|= tmp_logflags
;
5704 if (!isrt
&& wasdel
)
5705 xfs_mod_fdblocks(mp
, (int64_t)del
.br_blockcount
, false);
5707 bno
= del
.br_startoff
- 1;
5710 * If not done go on to the next (previous) record.
5712 if (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
) {
5714 xfs_iext_get_extent(ifp
, lastx
, &got
);
5715 if (got
.br_startoff
> bno
&& --lastx
>= 0)
5716 xfs_iext_get_extent(ifp
, lastx
, &got
);
5721 if (bno
== (xfs_fileoff_t
)-1 || bno
< start
|| lastx
< 0)
5724 *rlen
= bno
- start
+ 1;
5727 * Convert to a btree if necessary.
5729 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
5730 ASSERT(cur
== NULL
);
5731 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, dfops
,
5732 &cur
, 0, &tmp_logflags
, whichfork
);
5733 logflags
|= tmp_logflags
;
5738 * transform from btree to extents, give it cur
5740 else if (xfs_bmap_wants_extents(ip
, whichfork
)) {
5741 ASSERT(cur
!= NULL
);
5742 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &tmp_logflags
,
5744 logflags
|= tmp_logflags
;
5749 * transform from extents to local?
5754 * Log everything. Do this after conversion, there's no point in
5755 * logging the extent records if we've converted to btree format.
5757 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5758 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5759 logflags
&= ~xfs_ilog_fext(whichfork
);
5760 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5761 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5762 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5764 * Log inode even in the error case, if the transaction
5765 * is dirty we'll need to shut down the filesystem.
5768 xfs_trans_log_inode(tp
, ip
, logflags
);
5771 *firstblock
= cur
->bc_private
.b
.firstblock
;
5772 cur
->bc_private
.b
.allocated
= 0;
5774 xfs_btree_del_cursor(cur
,
5775 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5780 /* Unmap a range of a file. */
5784 struct xfs_inode
*ip
,
5789 xfs_fsblock_t
*firstblock
,
5790 struct xfs_defer_ops
*dfops
,
5795 error
= __xfs_bunmapi(tp
, ip
, bno
, &len
, flags
, nexts
, firstblock
,
5802 * Determine whether an extent shift can be accomplished by a merge with the
5803 * extent that precedes the target hole of the shift.
5807 struct xfs_bmbt_irec
*left
, /* preceding extent */
5808 struct xfs_bmbt_irec
*got
, /* current extent to shift */
5809 xfs_fileoff_t shift
) /* shift fsb */
5811 xfs_fileoff_t startoff
;
5813 startoff
= got
->br_startoff
- shift
;
5816 * The extent, once shifted, must be adjacent in-file and on-disk with
5817 * the preceding extent.
5819 if ((left
->br_startoff
+ left
->br_blockcount
!= startoff
) ||
5820 (left
->br_startblock
+ left
->br_blockcount
!= got
->br_startblock
) ||
5821 (left
->br_state
!= got
->br_state
) ||
5822 (left
->br_blockcount
+ got
->br_blockcount
> MAXEXTLEN
))
5829 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5830 * hole in the file. If an extent shift would result in the extent being fully
5831 * adjacent to the extent that currently precedes the hole, we can merge with
5832 * the preceding extent rather than do the shift.
5834 * This function assumes the caller has verified a shift-by-merge is possible
5835 * with the provided extents via xfs_bmse_can_merge().
5839 struct xfs_inode
*ip
,
5841 xfs_fileoff_t shift
, /* shift fsb */
5842 int current_ext
, /* idx of gotp */
5843 struct xfs_bmbt_rec_host
*gotp
, /* extent to shift */
5844 struct xfs_bmbt_rec_host
*leftp
, /* preceding extent */
5845 struct xfs_btree_cur
*cur
,
5846 int *logflags
) /* output */
5848 struct xfs_bmbt_irec got
;
5849 struct xfs_bmbt_irec left
;
5850 xfs_filblks_t blockcount
;
5852 struct xfs_mount
*mp
= ip
->i_mount
;
5854 xfs_bmbt_get_all(gotp
, &got
);
5855 xfs_bmbt_get_all(leftp
, &left
);
5856 blockcount
= left
.br_blockcount
+ got
.br_blockcount
;
5858 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5859 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5860 ASSERT(xfs_bmse_can_merge(&left
, &got
, shift
));
5863 * Merge the in-core extents. Note that the host record pointers and
5864 * current_ext index are invalid once the extent has been removed via
5865 * xfs_iext_remove().
5867 xfs_bmbt_set_blockcount(leftp
, blockcount
);
5868 xfs_iext_remove(ip
, current_ext
, 1, 0);
5871 * Update the on-disk extent count, the btree if necessary and log the
5874 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5875 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
5876 *logflags
|= XFS_ILOG_CORE
;
5878 *logflags
|= XFS_ILOG_DEXT
;
5882 /* lookup and remove the extent to merge */
5883 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
, got
.br_startblock
,
5884 got
.br_blockcount
, &i
);
5887 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5889 error
= xfs_btree_delete(cur
, &i
);
5892 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5894 /* lookup and update size of the previous extent */
5895 error
= xfs_bmbt_lookup_eq(cur
, left
.br_startoff
, left
.br_startblock
,
5896 left
.br_blockcount
, &i
);
5899 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5901 left
.br_blockcount
= blockcount
;
5903 return xfs_bmbt_update(cur
, left
.br_startoff
, left
.br_startblock
,
5904 left
.br_blockcount
, left
.br_state
);
5908 * Shift a single extent.
5912 struct xfs_inode
*ip
,
5914 xfs_fileoff_t offset_shift_fsb
,
5916 struct xfs_bmbt_rec_host
*gotp
,
5917 struct xfs_btree_cur
*cur
,
5919 enum shift_direction direction
,
5920 struct xfs_defer_ops
*dfops
)
5922 struct xfs_ifork
*ifp
;
5923 struct xfs_mount
*mp
;
5924 xfs_fileoff_t startoff
;
5925 struct xfs_bmbt_rec_host
*adj_irecp
;
5926 struct xfs_bmbt_irec got
;
5927 struct xfs_bmbt_irec adj_irec
;
5933 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5934 total_extents
= xfs_iext_count(ifp
);
5936 xfs_bmbt_get_all(gotp
, &got
);
5938 /* delalloc extents should be prevented by caller */
5939 XFS_WANT_CORRUPTED_RETURN(mp
, !isnullstartblock(got
.br_startblock
));
5941 if (direction
== SHIFT_LEFT
) {
5942 startoff
= got
.br_startoff
- offset_shift_fsb
;
5945 * Check for merge if we've got an extent to the left,
5946 * otherwise make sure there's enough room at the start
5947 * of the file for the shift.
5949 if (!*current_ext
) {
5950 if (got
.br_startoff
< offset_shift_fsb
)
5952 goto update_current_ext
;
5955 * grab the left extent and check for a large
5958 adj_irecp
= xfs_iext_get_ext(ifp
, *current_ext
- 1);
5959 xfs_bmbt_get_all(adj_irecp
, &adj_irec
);
5962 adj_irec
.br_startoff
+ adj_irec
.br_blockcount
)
5965 /* check whether to merge the extent or shift it down */
5966 if (xfs_bmse_can_merge(&adj_irec
, &got
,
5967 offset_shift_fsb
)) {
5968 error
= xfs_bmse_merge(ip
, whichfork
, offset_shift_fsb
,
5969 *current_ext
, gotp
, adj_irecp
,
5977 startoff
= got
.br_startoff
+ offset_shift_fsb
;
5978 /* nothing to move if this is the last extent */
5979 if (*current_ext
>= (total_extents
- 1))
5980 goto update_current_ext
;
5982 * If this is not the last extent in the file, make sure there
5983 * is enough room between current extent and next extent for
5984 * accommodating the shift.
5986 adj_irecp
= xfs_iext_get_ext(ifp
, *current_ext
+ 1);
5987 xfs_bmbt_get_all(adj_irecp
, &adj_irec
);
5988 if (startoff
+ got
.br_blockcount
> adj_irec
.br_startoff
)
5991 * Unlike a left shift (which involves a hole punch),
5992 * a right shift does not modify extent neighbors
5993 * in any way. We should never find mergeable extents
5994 * in this scenario. Check anyways and warn if we
5995 * encounter two extents that could be one.
5997 if (xfs_bmse_can_merge(&got
, &adj_irec
, offset_shift_fsb
))
6001 * Increment the extent index for the next iteration, update the start
6002 * offset of the in-core extent and update the btree if applicable.
6005 if (direction
== SHIFT_LEFT
)
6009 xfs_bmbt_set_startoff(gotp
, startoff
);
6010 *logflags
|= XFS_ILOG_CORE
;
6013 *logflags
|= XFS_ILOG_DEXT
;
6017 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
, got
.br_startblock
,
6018 got
.br_blockcount
, &i
);
6021 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
6023 got
.br_startoff
= startoff
;
6024 error
= xfs_bmbt_update(cur
, got
.br_startoff
, got
.br_startblock
,
6025 got
.br_blockcount
, got
.br_state
);
6030 /* update reverse mapping */
6031 error
= xfs_rmap_unmap_extent(mp
, dfops
, ip
, whichfork
, &adj_irec
);
6034 adj_irec
.br_startoff
= startoff
;
6035 return xfs_rmap_map_extent(mp
, dfops
, ip
, whichfork
, &adj_irec
);
6039 * Shift extent records to the left/right to cover/create a hole.
6041 * The maximum number of extents to be shifted in a single operation is
6042 * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6043 * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6044 * is the length by which each extent is shifted. If there is no hole to shift
6045 * the extents into, this will be considered invalid operation and we abort
6049 xfs_bmap_shift_extents(
6050 struct xfs_trans
*tp
,
6051 struct xfs_inode
*ip
,
6052 xfs_fileoff_t
*next_fsb
,
6053 xfs_fileoff_t offset_shift_fsb
,
6055 xfs_fileoff_t stop_fsb
,
6056 xfs_fsblock_t
*firstblock
,
6057 struct xfs_defer_ops
*dfops
,
6058 enum shift_direction direction
,
6061 struct xfs_btree_cur
*cur
= NULL
;
6062 struct xfs_bmbt_rec_host
*gotp
;
6063 struct xfs_bmbt_irec got
;
6064 struct xfs_mount
*mp
= ip
->i_mount
;
6065 struct xfs_ifork
*ifp
;
6066 xfs_extnum_t nexts
= 0;
6067 xfs_extnum_t current_ext
;
6068 xfs_extnum_t total_extents
;
6069 xfs_extnum_t stop_extent
;
6071 int whichfork
= XFS_DATA_FORK
;
6074 if (unlikely(XFS_TEST_ERROR(
6075 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
6076 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
6077 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
6078 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6079 XFS_ERRLEVEL_LOW
, mp
);
6080 return -EFSCORRUPTED
;
6083 if (XFS_FORCED_SHUTDOWN(mp
))
6086 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
6087 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
6088 ASSERT(direction
== SHIFT_LEFT
|| direction
== SHIFT_RIGHT
);
6089 ASSERT(*next_fsb
!= NULLFSBLOCK
|| direction
== SHIFT_RIGHT
);
6091 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6092 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
6093 /* Read in all the extents */
6094 error
= xfs_iread_extents(tp
, ip
, whichfork
);
6099 if (ifp
->if_flags
& XFS_IFBROOT
) {
6100 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
6101 cur
->bc_private
.b
.firstblock
= *firstblock
;
6102 cur
->bc_private
.b
.dfops
= dfops
;
6103 cur
->bc_private
.b
.flags
= 0;
6107 * There may be delalloc extents in the data fork before the range we
6108 * are collapsing out, so we cannot use the count of real extents here.
6109 * Instead we have to calculate it from the incore fork.
6111 total_extents
= xfs_iext_count(ifp
);
6112 if (total_extents
== 0) {
6118 * In case of first right shift, we need to initialize next_fsb
6120 if (*next_fsb
== NULLFSBLOCK
) {
6121 gotp
= xfs_iext_get_ext(ifp
, total_extents
- 1);
6122 xfs_bmbt_get_all(gotp
, &got
);
6123 *next_fsb
= got
.br_startoff
;
6124 if (stop_fsb
> *next_fsb
) {
6130 /* Lookup the extent index at which we have to stop */
6131 if (direction
== SHIFT_RIGHT
) {
6132 gotp
= xfs_iext_bno_to_ext(ifp
, stop_fsb
, &stop_extent
);
6133 /* Make stop_extent exclusive of shift range */
6136 stop_extent
= total_extents
;
6139 * Look up the extent index for the fsb where we start shifting. We can
6140 * henceforth iterate with current_ext as extent list changes are locked
6143 * gotp can be null in 2 cases: 1) if there are no extents or 2)
6144 * *next_fsb lies in a hole beyond which there are no extents. Either
6147 gotp
= xfs_iext_bno_to_ext(ifp
, *next_fsb
, ¤t_ext
);
6153 /* some sanity checking before we finally start shifting extents */
6154 if ((direction
== SHIFT_LEFT
&& current_ext
>= stop_extent
) ||
6155 (direction
== SHIFT_RIGHT
&& current_ext
<= stop_extent
)) {
6160 while (nexts
++ < num_exts
) {
6161 error
= xfs_bmse_shift_one(ip
, whichfork
, offset_shift_fsb
,
6162 ¤t_ext
, gotp
, cur
, &logflags
,
6167 * If there was an extent merge during the shift, the extent
6168 * count can change. Update the total and grade the next record.
6170 if (direction
== SHIFT_LEFT
) {
6171 total_extents
= xfs_iext_count(ifp
);
6172 stop_extent
= total_extents
;
6175 if (current_ext
== stop_extent
) {
6177 *next_fsb
= NULLFSBLOCK
;
6180 gotp
= xfs_iext_get_ext(ifp
, current_ext
);
6184 xfs_bmbt_get_all(gotp
, &got
);
6185 *next_fsb
= got
.br_startoff
;
6190 xfs_btree_del_cursor(cur
,
6191 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
6194 xfs_trans_log_inode(tp
, ip
, logflags
);
6200 * Splits an extent into two extents at split_fsb block such that it is
6201 * the first block of the current_ext. @current_ext is a target extent
6202 * to be split. @split_fsb is a block where the extents is split.
6203 * If split_fsb lies in a hole or the first block of extents, just return 0.
6206 xfs_bmap_split_extent_at(
6207 struct xfs_trans
*tp
,
6208 struct xfs_inode
*ip
,
6209 xfs_fileoff_t split_fsb
,
6210 xfs_fsblock_t
*firstfsb
,
6211 struct xfs_defer_ops
*dfops
)
6213 int whichfork
= XFS_DATA_FORK
;
6214 struct xfs_btree_cur
*cur
= NULL
;
6215 struct xfs_bmbt_rec_host
*gotp
;
6216 struct xfs_bmbt_irec got
;
6217 struct xfs_bmbt_irec
new; /* split extent */
6218 struct xfs_mount
*mp
= ip
->i_mount
;
6219 struct xfs_ifork
*ifp
;
6220 xfs_fsblock_t gotblkcnt
; /* new block count for got */
6221 xfs_extnum_t current_ext
;
6226 if (unlikely(XFS_TEST_ERROR(
6227 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
6228 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
6229 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
6230 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6231 XFS_ERRLEVEL_LOW
, mp
);
6232 return -EFSCORRUPTED
;
6235 if (XFS_FORCED_SHUTDOWN(mp
))
6238 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6239 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
6240 /* Read in all the extents */
6241 error
= xfs_iread_extents(tp
, ip
, whichfork
);
6247 * gotp can be null in 2 cases: 1) if there are no extents
6248 * or 2) split_fsb lies in a hole beyond which there are
6249 * no extents. Either way, we are done.
6251 gotp
= xfs_iext_bno_to_ext(ifp
, split_fsb
, ¤t_ext
);
6255 xfs_bmbt_get_all(gotp
, &got
);
6258 * Check split_fsb lies in a hole or the start boundary offset
6261 if (got
.br_startoff
>= split_fsb
)
6264 gotblkcnt
= split_fsb
- got
.br_startoff
;
6265 new.br_startoff
= split_fsb
;
6266 new.br_startblock
= got
.br_startblock
+ gotblkcnt
;
6267 new.br_blockcount
= got
.br_blockcount
- gotblkcnt
;
6268 new.br_state
= got
.br_state
;
6270 if (ifp
->if_flags
& XFS_IFBROOT
) {
6271 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
6272 cur
->bc_private
.b
.firstblock
= *firstfsb
;
6273 cur
->bc_private
.b
.dfops
= dfops
;
6274 cur
->bc_private
.b
.flags
= 0;
6275 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
6281 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, del_cursor
);
6284 xfs_bmbt_set_blockcount(gotp
, gotblkcnt
);
6285 got
.br_blockcount
= gotblkcnt
;
6287 logflags
= XFS_ILOG_CORE
;
6289 error
= xfs_bmbt_update(cur
, got
.br_startoff
,
6296 logflags
|= XFS_ILOG_DEXT
;
6298 /* Add new extent */
6300 xfs_iext_insert(ip
, current_ext
, 1, &new, 0);
6301 XFS_IFORK_NEXT_SET(ip
, whichfork
,
6302 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
6305 error
= xfs_bmbt_lookup_eq(cur
, new.br_startoff
,
6306 new.br_startblock
, new.br_blockcount
,
6310 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, del_cursor
);
6311 cur
->bc_rec
.b
.br_state
= new.br_state
;
6313 error
= xfs_btree_insert(cur
, &i
);
6316 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, del_cursor
);
6320 * Convert to a btree if necessary.
6322 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
6323 int tmp_logflags
; /* partial log flag return val */
6325 ASSERT(cur
== NULL
);
6326 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstfsb
, dfops
,
6327 &cur
, 0, &tmp_logflags
, whichfork
);
6328 logflags
|= tmp_logflags
;
6333 cur
->bc_private
.b
.allocated
= 0;
6334 xfs_btree_del_cursor(cur
,
6335 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
6339 xfs_trans_log_inode(tp
, ip
, logflags
);
6344 xfs_bmap_split_extent(
6345 struct xfs_inode
*ip
,
6346 xfs_fileoff_t split_fsb
)
6348 struct xfs_mount
*mp
= ip
->i_mount
;
6349 struct xfs_trans
*tp
;
6350 struct xfs_defer_ops dfops
;
6351 xfs_fsblock_t firstfsb
;
6354 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
,
6355 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0, 0, &tp
);
6359 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
6360 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
6362 xfs_defer_init(&dfops
, &firstfsb
);
6364 error
= xfs_bmap_split_extent_at(tp
, ip
, split_fsb
,
6369 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
6373 return xfs_trans_commit(tp
);
6376 xfs_defer_cancel(&dfops
);
6377 xfs_trans_cancel(tp
);
6381 /* Deferred mapping is only for real extents in the data fork. */
6383 xfs_bmap_is_update_needed(
6384 struct xfs_bmbt_irec
*bmap
)
6386 return bmap
->br_startblock
!= HOLESTARTBLOCK
&&
6387 bmap
->br_startblock
!= DELAYSTARTBLOCK
;
6390 /* Record a bmap intent. */
6393 struct xfs_mount
*mp
,
6394 struct xfs_defer_ops
*dfops
,
6395 enum xfs_bmap_intent_type type
,
6396 struct xfs_inode
*ip
,
6398 struct xfs_bmbt_irec
*bmap
)
6401 struct xfs_bmap_intent
*bi
;
6403 trace_xfs_bmap_defer(mp
,
6404 XFS_FSB_TO_AGNO(mp
, bmap
->br_startblock
),
6406 XFS_FSB_TO_AGBNO(mp
, bmap
->br_startblock
),
6407 ip
->i_ino
, whichfork
,
6409 bmap
->br_blockcount
,
6412 bi
= kmem_alloc(sizeof(struct xfs_bmap_intent
), KM_SLEEP
| KM_NOFS
);
6413 INIT_LIST_HEAD(&bi
->bi_list
);
6416 bi
->bi_whichfork
= whichfork
;
6417 bi
->bi_bmap
= *bmap
;
6419 error
= xfs_defer_join(dfops
, bi
->bi_owner
);
6425 xfs_defer_add(dfops
, XFS_DEFER_OPS_TYPE_BMAP
, &bi
->bi_list
);
6429 /* Map an extent into a file. */
6431 xfs_bmap_map_extent(
6432 struct xfs_mount
*mp
,
6433 struct xfs_defer_ops
*dfops
,
6434 struct xfs_inode
*ip
,
6435 struct xfs_bmbt_irec
*PREV
)
6437 if (!xfs_bmap_is_update_needed(PREV
))
6440 return __xfs_bmap_add(mp
, dfops
, XFS_BMAP_MAP
, ip
,
6441 XFS_DATA_FORK
, PREV
);
6444 /* Unmap an extent out of a file. */
6446 xfs_bmap_unmap_extent(
6447 struct xfs_mount
*mp
,
6448 struct xfs_defer_ops
*dfops
,
6449 struct xfs_inode
*ip
,
6450 struct xfs_bmbt_irec
*PREV
)
6452 if (!xfs_bmap_is_update_needed(PREV
))
6455 return __xfs_bmap_add(mp
, dfops
, XFS_BMAP_UNMAP
, ip
,
6456 XFS_DATA_FORK
, PREV
);
6460 * Process one of the deferred bmap operations. We pass back the
6461 * btree cursor to maintain our lock on the bmapbt between calls.
6464 xfs_bmap_finish_one(
6465 struct xfs_trans
*tp
,
6466 struct xfs_defer_ops
*dfops
,
6467 struct xfs_inode
*ip
,
6468 enum xfs_bmap_intent_type type
,
6470 xfs_fileoff_t startoff
,
6471 xfs_fsblock_t startblock
,
6472 xfs_filblks_t blockcount
,
6475 struct xfs_bmbt_irec bmap
;
6477 xfs_fsblock_t firstfsb
;
6478 int flags
= XFS_BMAPI_REMAP
;
6482 bmap
.br_startblock
= startblock
;
6483 bmap
.br_startoff
= startoff
;
6484 bmap
.br_blockcount
= blockcount
;
6485 bmap
.br_state
= state
;
6487 trace_xfs_bmap_deferred(tp
->t_mountp
,
6488 XFS_FSB_TO_AGNO(tp
->t_mountp
, startblock
), type
,
6489 XFS_FSB_TO_AGBNO(tp
->t_mountp
, startblock
),
6490 ip
->i_ino
, whichfork
, startoff
, blockcount
, state
);
6492 if (whichfork
!= XFS_DATA_FORK
&& whichfork
!= XFS_ATTR_FORK
)
6493 return -EFSCORRUPTED
;
6494 if (whichfork
== XFS_ATTR_FORK
)
6495 flags
|= XFS_BMAPI_ATTRFORK
;
6497 if (XFS_TEST_ERROR(false, tp
->t_mountp
,
6498 XFS_ERRTAG_BMAP_FINISH_ONE
,
6499 XFS_RANDOM_BMAP_FINISH_ONE
))
6504 firstfsb
= bmap
.br_startblock
;
6505 error
= xfs_bmapi_write(tp
, ip
, bmap
.br_startoff
,
6506 bmap
.br_blockcount
, flags
, &firstfsb
,
6507 bmap
.br_blockcount
, &bmap
, &nimaps
,
6510 case XFS_BMAP_UNMAP
:
6511 error
= xfs_bunmapi(tp
, ip
, bmap
.br_startoff
,
6512 bmap
.br_blockcount
, flags
, 1, &firstfsb
,
6518 error
= -EFSCORRUPTED
;